# ModelPiper > ModelPiper is a suite of local-first AI apps for macOS. A browser-based visual pipeline builder connects to native companion apps that handle LLM inference, speech, vision, OCR, NLP, RAG, image/video upscaling, browser automation, and more, all running on-device with no cloud dependency. ## Products - [ModelPiper Web App](https://modelpiper.com/pipeline): Browser-based visual drag-and-drop AI pipeline builder. Block-based interface for connecting LLMs, vision, TTS, STT, and image models into multi-step workflows. Chat and image processing are free to use; video upscaling requires ToolPiper Studio ($29/mo) and E2E test authoring requires Max ($49/mo). - [ToolPiper for macOS](https://modelpiper.com/download): Local AI gateway and MCP server. Bundles 9 inference backends (llama.cpp, Apple Intelligence, FluidAudio STT, FluidAudio TTS, MLX Audio TTS, Apple Vision OCR, CoreML Image Upscale, CoreML Pose, CoreML Embedding). 421 MCP tools across 14 categories: core inference, analysis, browser, filesystem, capture, media, testing, motion, outreach, search, sieve, system, video, and OAuth. Includes 26 system-action domains (audio, display, network, Bluetooth, calendar, contacts, Finder, notifications, power, process, reminders, system settings, window management) with 142 actions and push-to-talk voice commands (140ms STT on Apple Neural Engine). OpenAI-compatible API on localhost. Cloud API proxy with Keychain key injection. Free DMG download. Paid rungs: Pro ($10/mo) unlocks local RAG, web scraping with YouTube transcripts, and the cloud API proxy; Studio ($29/mo) adds image and video upscaling, the video editing pipeline, pose detection, and the outreach toolkit; Max ($49/mo) adds the agentic code editor, PiperTest, and API discovery. Team is $99/mo per deployment with unlimited named member tokens. - [VisionPiper for macOS](https://modelpiper.com/visionpiper): Screen capture, MP4 recording, GIF conversion, and live WebSocket streaming. Region selection with pass-through interior. Free on the Mac App Store. - [AudioPiper for macOS](https://modelpiper.com/audiopiper): Multi-source audio mixer and recorder. Captures microphone, system audio, and per-app audio via Core Audio Taps. Actor-based streaming WAV recording. Free on the Mac App Store. - [MediaPiper Browser Extension](https://chromewebstore.google.com/detail/mediapiper/gknkckkachhflfmhdiaeaomacceechnb): Chrome, Firefox, and Safari extension for full-size image preview on hover. Declarative sieve engine resolves thumbnails to full-size images. Free to install; AI image upscaling runs through ToolPiper and requires Studio ($29/mo). ## Documentation - [Homepage](https://modelpiper.com/): Product overview, features, pricing, and getting started guide - [Documentation Hub](https://modelpiper.com/docs): All docs: 16 capabilities, 5 apps - [ToolPiper API Reference](https://modelpiper.com/docs/toolpiper): REST API documentation, 90+ endpoints, dev tokens - [MCP Server Docs](https://modelpiper.com/docs/mcp): 421 MCP tools, stdio + HTTP transports, setup for Claude Code / Cursor / Windsurf - [MCP Landing Page](https://modelpiper.com/mcp): MCP tool categories, installation guide, per-feature filtering - [MCP Tool Catalog](https://modelpiper.com/mcp/tools): Searchable catalog of all 421 built-in MCP tools: name, one-line description, category, and unlock tier for every tool - [MCP Tools Marketplace](https://modelpiper.com/mcp-tools): Curated third-party MCP servers installable in ToolPiper with one click. 51 servers across developer tools, data, productivity, search, content, system, and AI tooling - [GitHub MCP Server](https://modelpiper.com/mcp-tools/github): Create issues, manage pull requests, and search repositories from your AI assistant - [Filesystem MCP Server](https://modelpiper.com/mcp-tools/filesystem): Read and write files in a directory you choose - [Postgres MCP Server](https://modelpiper.com/mcp-tools/postgres): Run SQL queries against a Postgres database - [Slack MCP Server](https://modelpiper.com/mcp-tools/slack): Read channels, post messages, and search Slack workspaces - [Fetch MCP Server](https://modelpiper.com/mcp-tools/fetch): Fetch URLs and parse the content into markdown for the model - [Git MCP Server](https://modelpiper.com/mcp-tools/git): Read, search, and manipulate Git repositories on disk - [Memory MCP Server](https://modelpiper.com/mcp-tools/memory): Knowledge-graph backed persistent memory across chats - [Sequential Thinking MCP Server](https://modelpiper.com/mcp-tools/sequential-thinking): Dynamic, reflective problem-solving through thought sequences - [Time MCP Server](https://modelpiper.com/mcp-tools/time): Time and timezone conversion utilities - [Everything MCP Server](https://modelpiper.com/mcp-tools/everything): Reference test server exercising every MCP feature - [SQLite MCP Server](https://modelpiper.com/mcp-tools/sqlite): Query SQLite databases and run business-intelligence prompts - [Redis MCP Server](https://modelpiper.com/mcp-tools/redis): Interact with Redis key-value stores from the assistant - [Puppeteer MCP Server](https://modelpiper.com/mcp-tools/puppeteer): Headless browser automation and web scraping - [GitLab MCP Server](https://modelpiper.com/mcp-tools/gitlab): Manage GitLab projects, issues, and merge requests - [Google Drive MCP Server](https://modelpiper.com/mcp-tools/google-drive): Search and read files from Google Drive - [Google Maps MCP Server](https://modelpiper.com/mcp-tools/google-maps): Geocoding, directions, and place lookup via Google Maps - [Brave Search MCP Server](https://modelpiper.com/mcp-tools/brave-search): Web and local search via Brave's privacy-focused search API - [Sentry MCP Server](https://modelpiper.com/mcp-tools/sentry): Retrieve and analyze issues from Sentry.io - [AWS Knowledge Base MCP Server](https://modelpiper.com/mcp-tools/aws-kb-retrieval): Retrieve answers from an AWS Bedrock Knowledge Base - [EverArt MCP Server](https://modelpiper.com/mcp-tools/everart): AI image generation using multiple model backends - [ModelPiper Web App Docs](https://modelpiper.com/docs/modelpiper): Pipeline builder, block types, providers, workflows - [VisionPiper Docs](https://modelpiper.com/docs/visionpiper): Screen capture, recording, streaming - [AudioPiper Docs](https://modelpiper.com/docs/audiopiper): Audio mixer, per-app capture, recording - [MediaPiper Docs](https://modelpiper.com/docs/mediapiper): Browser extension, sieve engine, AI upscale - [Blog](https://modelpiper.com/blog): 152 technical articles and papers - [Blog RSS Feed](https://modelpiper.com/blog/feed.xml): RSS 2.0 feed of all blog articles and papers - [Workflows](https://modelpiper.com/workflow): 9 definitive guides to local AI workflows on Mac - [Changelog](https://modelpiper.com/changelog): Versioned release notes, also available as Markdown at /changelog.md - [OpenAPI Specification](https://modelpiper.com/api/openapi.json): Machine-readable OpenAPI 3.1 spec for the ToolPiper REST API (200+ operations) - [API Catalog](https://modelpiper.com/.well-known/api-catalog): RFC 9727 linkset pointing at the spec and human-facing docs ## Model Fit and Mac Hardware Which local AI models a given Mac can actually run, from measured memory bandwidth and unified memory ceilings rather than vendor claims. - [Model Fit Checker](https://modelpiper.com/fit): Pick a Mac, see which of 3,641 open models fit, with quantization recommendations, memory requirements, and estimated tokens/sec. - [Benchmark Leaderboard](https://modelpiper.com/fit/benchmarks): How prompt-processing and token-generation rates are measured by chip, quantization, and backend. Community results are not published yet; the per-model tok/s figures on /fit pages are formula estimates and are labelled as such. ### Per-model pages The catalog carries **3,641 models, one page each**, at `/fit/`. Every page answers the same question for one model: which Macs run it, at which quantization, with what memory headroom and what tokens/sec. Slugs are derived from the HuggingFace repo id: lowercased, with `/` and `.` both replaced by `-`. So `Qwen/Qwen2.5-7B-Instruct` becomes `/fit/qwen-qwen2-5-7b-instruct`, and any model id in the catalog can be turned into its URL by that rule without looking it up. Each page carries: parameter count, architecture, context length, license, per-quantization memory requirements (Q4 through F16), a fit verdict for every Apple Silicon chip, and estimated generation speed. A representative head, chosen for coverage of size and task rather than raw download rank (which is dominated by sentence-embedding models): - [Qwen3 0.6B](https://modelpiper.com/fit/qwen-qwen3-0-6b): 0.75B general, runs on every Apple Silicon Mac including 8 GB - [Qwen3 4B](https://modelpiper.com/fit/qwen-qwen3-4b): 4B general - [Mistral 7B Instruct v0.3](https://modelpiper.com/fit/mistralai-mistral-7b-instruct-v0-3): 7.25B chat - [Qwen2.5 7B Instruct](https://modelpiper.com/fit/qwen-qwen2-5-7b-instruct): 7.62B chat - [Llama 3.1 8B Instruct](https://modelpiper.com/fit/meta-llama-llama-3-1-8b-instruct): 8.03B chat - [Qwen3 8B](https://modelpiper.com/fit/qwen-qwen3-8b): 8.19B general - [Qwen2.5 VL 7B Instruct](https://modelpiper.com/fit/qwen-qwen2-5-vl-7b-instruct): 8.29B vision-language - [Qwen3.6 27B](https://modelpiper.com/fit/qwen-qwen3-6-27b): 27.78B multimodal - [Qwen3 30B A3B](https://modelpiper.com/fit/qwen-qwen3-30b-a3b): 30.53B mixture-of-experts - [Gemma 4 31B IT](https://modelpiper.com/fit/google-gemma-4-31b-it): 31B multimodal - [DeepSeek R1 Distill Qwen 32B](https://modelpiper.com/fit/deepseek-ai-deepseek-r1-distill-qwen-32b): 32.76B reasoning - [Qwen2.5 Coder 32B Instruct](https://modelpiper.com/fit/qwen-qwen2-5-coder-32b-instruct): 32.51B coding - [Qwen3 32B](https://modelpiper.com/fit/qwen-qwen3-32b): 32.76B general - [Llama 3.3 70B Instruct](https://modelpiper.com/fit/meta-llama-llama-3-3-70b-instruct): 70.55B chat, needs 64 GB unified memory at Q4 - [GPT-OSS 120B](https://modelpiper.com/fit/openai-gpt-oss-120b): 120.41B general, Mac Studio territory - [MacBook Air](https://modelpiper.com/macbook-air): Which LLMs run on a fanless MacBook Air, M1 through M5. - [MacBook Pro 14"](https://modelpiper.com/macbook-pro-14): Base, Pro, and Max chips compared by memory bandwidth and unified memory. - [MacBook Pro 16"](https://modelpiper.com/macbook-pro-16): Pro and Max chips, M1 through M5, ranked for local inference. - [Mac mini](https://modelpiper.com/mac-mini): The cheapest way into local AI on Apple Silicon. - [Mac Studio](https://modelpiper.com/mac-studio): Max and Ultra chips, the highest memory bandwidth and unified memory Apple ships, up to 512 GB. - [iMac](https://modelpiper.com/imac): Base M chips, where unified memory is the limit that matters. ### Per-chip pages Each of the six products above also has one page per chip it ships, at `//`, **42 pages** in total. The chip slug lowercases the Apple chip name and replaces spaces with `-`, so `M3 Ultra` under Mac Studio is `/mac-studio/m3-ultra`. Worked examples: - [Mac Studio M3 Ultra](https://modelpiper.com/mac-studio/m3-ultra): highest memory bandwidth and unified memory Apple ships - [MacBook Pro 14" M4 Max](https://modelpiper.com/macbook-pro-14/m4-max): the portable ceiling for local inference - [Mac mini M4 Pro](https://modelpiper.com/mac-mini/m4-pro): cheapest route to 64 GB unified memory Each chip page carries memory bandwidth, unified memory options, CPU/GPU core counts, Neural Engine TOPS, the largest dense model that chip holds at Q4, and a per-model fit table. ## Key Facts - ToolPiper is an MCP server with 421 tools across 14 categories and 5 resources, stdio and HTTP transports - 9 bundled inference backends: llama.cpp, Apple Intelligence, FluidAudio STT, FluidAudio TTS, MLX Audio TTS, Apple Vision OCR, CoreML Image Upscale, CoreML Pose, CoreML Embedding - PiperSR video upscale: 44.4 FPS on M4 Max (1.5x realtime) using double-buffered ANE+Metal pipeline - Custom CDP browser automation engine with AX-native selectors, self-healing (~5-15ms), and mutation diffing - PiperTest (Max, $49/mo): visual AX-native test format with Playwright/Cypress export - 16-framework detection in web scraper (React, Vue, Angular, Svelte, Next.js, Nuxt, etc.) - HNSW + BM25 hybrid RAG with semantic chunking, EmbeddingGemma embeddings on the Neural Engine (zero-setup) - Voice cloning via Qwen3 TTS with reference audio - Push-to-talk dictation: 140ms on Apple Neural Engine, 25 languages - ToolPiper system actions: 26 action domains, 142 actions, 181 MCP tools for desktop automation (built in to ToolPiper) - All AI processing runs locally on-device, no cloud required unless user opts in - ToolPiper requires macOS 26 or later, optimized for Apple Silicon - Privacy-focused: prompts, responses, and workflows never leave the device - All companion apps (VisionPiper, AudioPiper, MediaPiper) are free to install; MediaPiper's AI upscaling requires ToolPiper Studio ($29/mo) ## MCP Server Setup ``` claude mcp add toolpiper -- ~/.toolpiper/mcp ``` Or in .mcp.json: `{ "mcpServers": { "toolpiper": { "type": "stdio", "command": "~/.toolpiper/mcp" } } }` HTTP transport: `{ "mcpServers": { "toolpiper": { "type": "http", "url": "http://localhost:9998/mcp" } } }` ## Community - [GitHub Discussions](https://github.com/ModelPiper/modelpiper-community/discussions): Community discussion board - [Hugging Face](https://huggingface.co/ModelPiper): Published model weights, including PiperSR-2x - [X / Twitter](https://x.com/modelpiper) - [LinkedIn](https://www.linkedin.com/company/piperkit/) - [Reddit](https://www.reddit.com/r/ModelPiper/) ## Legal - [Privacy Policy](https://modelpiper.com/privacy) - [Terms of Service](https://modelpiper.com/terms) --- # Full Content > Concatenated Markdown bodies for every prerendered route. Source of truth: per-route `.md` twins emitted by `scripts/extract-content.mjs`. Regenerated on each build — do not hand-edit. ## Products URL: https://modelpiper.com/audiopiper # AudioPiper: Capture Any Audio on Your Mac > Per-app audio capture via Core Audio Taps. No virtual drivers, no kernel extensions, no Soundflower. Record or stream to AI, from any app. Free on the Mac App Store. **Free on Mac App Store** ## Capture any audio on your Mac. Per-app audio capture via Core Audio Taps. No virtual drivers, no kernel extensions, no Soundflower. Record or stream to AI, from any app. macOS 14.2 introduced Core Audio Taps, a proper Apple API for capturing audio from individual apps without installing system-level audio routing software. AudioPiper wraps this API in a free menu bar app that captures, mixes, and streams audio to ToolPiper for real-time transcription. - [Download from Mac App Store](https://apps.apple.com/us/app/audiopiper/id6759811522?mt=12) - [See voice AI workflows](/workflows/voice-ai) ## Per-App Audio Capture #### Capture Any App Zoom, Chrome, Spotify, Slack: capture audio from individual apps. Toggle sources on and off with one click. #### Mix Sources Combine microphone + system audio + per-app audio. Per-source volume controls let you balance levels in real time. #### No Virtual Drivers Uses Apple's Core Audio Tap API (macOS 14.2+). No Soundflower, BlackHole, or Loopback. Nothing to install, nothing to break on macOS updates. #### Constant Memory Actor-based streaming WAV recorder uses constant memory regardless of recording length. Record for 5 minutes or 5 hours, memory stays flat. #### Waveform Visualization Real-time waveform display with 8x gain for visual monitoring. See audio levels from every source at a glance. #### Background Capture Sits in the menu bar. Toggle sources with one click. No window to keep open, no UI in your way. ## Stream to AI #### Live Transcription Stream audio to FluidAudio for real-time speech-to-text as you speak. All processing runs on your Mac. #### Meeting Notes Capture Zoom audio, transcribe locally, summarize with an LLM, all on-device. No audio uploaded anywhere. #### Selective Capture Only capture the browser tab playing a webinar, not Spotify in the background. Per-app capture means you choose exactly what the AI hears. #### WebSocket Streaming Audio streams over WebSocket to ToolPiper on port 10002. Low-latency, real-time audio delivery for AI processing. ## What It Records #### Microphone input (built-in or external) #### System audio (all apps combined) #### Per-app audio via Core Audio Taps (macOS 14.2+) #### Mixed sources (mic + app audio simultaneously) #### WAV format output (lossless) #### DRM-protected audio (Apple Music, Netflix) #### Pre-selecting which apps to capture (you toggle after launch) #### Compressed formats like MP3 or AAC (records WAV only) ## FAQ ### Does it work without ToolPiper? Yes, for recording. AudioPiper records audio to WAV files independently, no other software required. You need ToolPiper for live AI transcription, where AudioPiper streams audio over WebSocket and ToolPiper runs speech-to-text on-device. ### Can it capture DRM audio? No. Core Audio Taps respect DRM restrictions enforced by macOS. Audio from DRM-protected sources like Apple Music or Netflix is not available for capture. This is an Apple-level restriction, not an AudioPiper limitation. ### What macOS version do I need? macOS 14.2 (Sonoma) or later for per-app audio capture via Core Audio Taps. Microphone and system audio capture work on macOS 13+. Apple Silicon (M1 or later) is required. ### What format does it record? WAV (lossless, uncompressed). This is ideal for AI processing and archival. Files are roughly 10 MB per minute of stereo audio at 48kHz. If you need MP3 or AAC, convert the WAV after recording with any audio tool. URL: https://modelpiper.com/pipertest # PiperTest: Self-Healing Browser Tests on Mac > Record browser tests visually using Chrome's real accessibility tree. Self-heal broken selectors automatically. Export to Playwright or Cypress. Local, no vendor lock-in. **Max** ## Browser tests that heal themselves. Record tests visually. Self-heal broken selectors. Export to Playwright or Cypress. All running locally on your Mac. End-to-end tests are the most valuable tests a team can write and the ones they write least. CSS selectors break on every refactor. Enterprise tools put self-healing behind a quote-only sales call. PiperTest queries Chrome's real accessibility tree, heals broken selectors in 5-15ms with zero external calls, and exports clean code for your CI pipeline. No cloud dependency, no seat-based pricing, no vendor lock-in. PiperTest is part of ToolPiper Max at $49/month. - [Download ToolPiper](https://modelpiper.com/download) - [Read the deep dive](/workflows/ai-testing) ## Why AX-Native Testing? Every major testing tool targets the DOM, which is a build artifact. PiperTest targets the accessibility tree, which represents what users actually see. #### Real Accessibility Tree PiperTest queries Chrome's native AX tree via CDP's Accessibility.queryAXTree. Not a DOM simulation, not injected JavaScript scanning every element. The actual computed accessibility tree that screen readers consume. #### Self-Healing Selectors Three healing modes: passive quality improvement upgrades weak selectors silently, AX fuzzy matching resolves breaks in 5-15ms locally, and AI-assisted healing handles the rest. Enterprise-grade healing, zero enterprise pricing. #### Framework-Proof A React-to-Vue migration doesn't touch the AX tree if the UI looks and behaves the same. CSS refactors, component library swaps, build tool upgrades, none of it breaks AX selectors. Tests break when behavior changes, which is exactly when they should. ## Everything You Need Full testing toolkit built on raw CDP. No wrappers, no abstractions, no compromises. #### Visual Recording Browse your app normally. Every interaction becomes an AX-enriched test step with element metadata, page context, and a mutation diff showing what changed. No annotation, no code, no switching between panels. #### 7 Assertion Types Visible, hidden, text content, URL match, element count, attribute value, and console message. All assertions use polling with configurable timeouts and capture an AX snapshot on failure for debugging. #### Temporal Assertions Three modes for time-dependent verification: always (condition holds for a duration), eventually (condition becomes true within a deadline), and next (condition holds on the very next check). No brittle waits. #### Health Monitors Passive console error, JavaScript exception, and HTTP error checking runs after every test step. Zero configuration. If your app throws a console error during a test, you'll know. #### Coverage Reports Three coverage dimensions in one weighted report: PiperProbe element coverage (60%), CDP JavaScript coverage (30%), and CSS coverage (10%). Color-coded bar with expandable uncovered elements per page. #### Playwright / Cypress Export One-click deterministic export. AX selectors map to each framework's native format: role:button:Sign In becomes page.getByRole('button', { name: 'Sign In' }) in Playwright. Clean, idiomatic code ready for CI. ## 20 MCP Tools for AI Agents Any MCP-capable AI client (Claude Code, Cursor, Windsurf) can create, run, heal, and export tests. #### browser_snapshot Returns Chrome's real AX tree as structured plain text. Auto-connects to Chrome on first call. The AI reads the page the same way a screen reader does: semantic roles, names, and states, not raw HTML. #### browser_action Click, fill, select, hover, scroll, and type using AX selectors. Self-healing resolves broken selectors automatically. Returns a structured AX diff showing exactly what changed on the page after the action. #### test_run Execute a saved PiperTest session with optional self-healing, health monitoring, temporal assertions, and coverage collection. Returns pass/fail summary with detailed results per step. #### test_export Export any PiperTest session to idiomatic Playwright or Cypress code. The AI can generate tests from a browser snapshot, save them, run them, heal failures, and export the final result, all through MCP tools. ## FAQ ### Does PiperTest replace Playwright? For test authoring and iteration, yes. PiperTest's visual recorder, self-healing, and inline editor handle the create-and-debug loop without code. For CI execution, no. PiperTest exports to Playwright (or Cypress) code that runs in your existing pipeline. Think of PiperTest as the authoring tool and Playwright as the CI runtime. ### Is PiperTest free? The browser automation underneath it is. The 14 browser MCP tools, the CDP connection, and Chrome's real accessibility tree are free on every tier, with no account. PiperTest itself is part of ToolPiper Max ($49/month): the 6 test tools (test\_save, test\_run, test\_list, test\_get, test\_export, test\_delete), sieve self-healing, visual recording, assertions, health monitoring, temporal assertions, coverage, and Playwright/Cypress export. ### How does self-healing work without AI? The default healing mode uses local AX fuzzy matching with zero external calls. When a selector fails, PiperTest takes a fresh AX tree snapshot and searches for nodes matching the original selector's role and approximate name. Candidates are scored by role match, name edit distance, and tree position. High-confidence matches execute automatically in 5-15ms. ### Can AI write tests for me? Yes, through MCP tools. Any MCP-capable AI client takes a browser snapshot (which returns the AX tree as plain text), reasons about what should be tested, generates PiperTest steps, saves them, runs them, and reports results. This works with Claude Code, Cursor, Windsurf, or any MCP client. ### Why AX selectors instead of data-testid? data-testid attributes are stable anchors, but they require developers to add them to every testable element, which means test infrastructure leaks into production code. AX selectors use what's already there: the element's role and accessible name. If your app is accessible, it's testable. If an AX selector doesn't resolve, it often means the element isn't accessible, which is a bug worth knowing about. URL: https://modelpiper.com/toolpiper # ToolPiper: Local AI Platform for Mac > Everything Ollama does, free: model downloads, the native llama.cpp engine, multi-model, a local OpenAI-compatible API. Then ToolPiper adds over 420 MCP tools, browser automation, voice, vision, and pipelines. **Free Direct Download** ## Everything Ollama does, free. Then it does the rest of your Mac. Model downloads, the native llama.cpp engine, multi-model, a local OpenAI-compatible API. Free, no account, no caps, no terminal. Then over 420 MCP tools, browser automation, voice, vision, and pipelines on top. Powered by llama.cpp: ToolPiper embeds upstream llama-server directly, not a fork, currently build b10068, with the exact version shown in the About panel. Models are standard GGUF files. One install replaces Ollama + Open WebUI + Playwright MCP + a filesystem MCP + LangChain. - [Download for Mac](https://modelpiper.com/download) - [Set up MCP](/blog/mcp-server-local-mac) ## Local AI Inference #### llama.cpp on Metal GPU Run any GGUF model. Qwen 3.5, Llama 3.2, DeepSeek R1. 30+ tok/s on M2 Air. Flash attention, speculative decoding, and Jinja templates out of the box. #### Apple Intelligence Neural Engine inference. Summarization, rewriting, Smart Reply. Zero GPU impact, runs on dedicated ML hardware alongside your other models. #### FluidAudio STT/TTS Whisper-class transcription at 210x realtime. PocketTTS with zero GPU impact. Both run on the Neural Engine, leaving Metal GPU free for LLM inference. #### MLX Audio TTS Soprano, Orpheus, voice cloning via Qwen3 TTS on Metal GPU. High-quality voice synthesis with support for multiple speakers and styles. #### Apple Vision OCR On-device text extraction from images and PDFs. No model download needed, uses the Vision framework built into macOS. #### Resource Intelligence Real-time RAM monitoring via proc_pid_rusage. Automatic model eviction under memory pressure. Your Mac stays responsive without manual intervention. ## Over 420 MCP Tools #### Core AI (33 tools) chat, audio_transcribe, audio_speak, audio_voice_clone, text_embed, vision_ocr, model_load/list/search/download, voice_chat memory, endpoint management. All inference on Neural Engine and Metal GPU. #### System Control (180 tools) 26 native macOS domains, all in-process via ActionRouter. Windows, displays, audio devices, network, Bluetooth, Dock, Spaces, Finder, calendar, contacts, focus, clipboard history, notifications, processes, reminders, timers. #### Browser & Web (32 tools) AX-native selectors, self-healing, 7 assertion types, visual recording, network interception, coverage, WebAuthn, autofill, plus web_scrape, http_request, youtube_transcript, and web_api_discover. #### Video Creator (65 tools) AI-driven screenplay-to-MP4 pipeline. screenplay, rehearse, record, render, narrate, plus full timeline edit (composition, narration, screenplay, timeline, export), import_media, and clip library. #### Filesystem & Git (24 tools) file_read, file_write, file_create, file_delete, file_list_directory, file_pick_directory, code_search, workspace_search, plus git_status, git_diff, git_commit, git_log, git_push, git_checkout. #### Outreach & Social (18 tools) github_repo_list, github_activity, github_compare, hn_search, hn_trending, reddit_search, reddit_post, gsc_analytics, gsc_inspect, queue_publish, queue_add, queue_list. Build your distribution loop. #### Analysis & RAG (48 tools) rag_ingest, rag_query, rag_collection_list with local embeddings + HNSW vector index. image_analyze, text_analyze, image_upscale, video_upscale, image_transform, pdf_extract, qr_generate. #### Capture & Motion (8 tools) vision_screenshot, vision_color_pick, audio_record, plus 60fps pose streaming: pose_detect, pose_format_list, pose_stream_start, vision_stream_start/stop. #### Testing & Integrations (14 tools) PiperTest (test_save, test_run, test_list, test_get, test_export, test_delete) with self-healing selectors. Plus OAuth (4), Sieve content filters (4) for end-to-end workflows. ## OpenAI-Compatible API #### /v1/chat/completions Drop-in replacement for OpenAI SDK. Change base_url to localhost:9998, that's it. Streaming and non-streaming. Works with LangChain, LlamaIndex, Continue.dev, Aider, and anything that accepts a custom OpenAI base URL. #### /v1/embeddings Local vector embeddings for RAG pipelines. EmbeddingGemma on the Neural Engine (zero-setup, 768-dim) or llama.cpp embedding models. Content-addressed cache for repeat queries. #### /v1/audio/speech Text-to-speech synthesis. Three engines (FluidAudio, MLX Audio, PocketTTS), eight voices, all on-device. Same API format as OpenAI's TTS endpoint. #### /v1/audio/transcriptions Speech-to-text transcription at 210x realtime on Neural Engine. Parakeet V3 model. Same API format as OpenAI's Whisper endpoint. #### Cloud API Proxy Route cloud requests through ToolPiper with Keychain key injection. Your API keys never appear in code or .env files. One base URL handles both local and cloud models. #### Developer Tokens tp_<64hex> Bearer tokens for team sharing and CI pipelines. SHA-256 hashed, stored in macOS Keychain. Works as api_key in any OpenAI SDK. ## Browser Automation & Testing #### AX-Native Selectors Query Chrome's real accessibility tree via CDP's Accessibility.queryAXTree. Not a DOM simulation, not injected JavaScript. The actual computed AX tree that screen readers consume. #### Self-Healing 3 modes: passive, fuzzy AX match, AI-assisted. Broken selectors repair in 5-15ms with zero external calls. Free, not a paid add-on. #### Visual Recording Browse your app normally. Every interaction becomes an AX-enriched test step with element metadata, page context, and a mutation diff showing what changed. #### Export to Playwright/Cypress Deterministic, idiomatic code for CI. AX selectors map to each framework's native format: role:button:Sign In becomes page.getByRole('button', { name: 'Sign In' }). ## Voice AI & Media #### Speech-to-Text Parakeet V3 at 210x realtime on Neural Engine. Whisper-class accuracy with zero GPU impact. Transcribe meetings, lectures, and voice memos locally. #### Text-to-Speech Three engines, eight voices, all on-device. FluidAudio on Neural Engine for low-latency. MLX Audio on Metal GPU for studio-quality synthesis. #### Voice Cloning Clone any voice from a 10-second sample via Qwen3 TTS on Metal GPU. Create custom voices for narration, accessibility, or creative projects. #### Video Upscale PiperSR at 44.4 FPS on Apple Neural Engine. Double-buffered ANE+Metal pipeline. Real-time 2x upscale from 360p to 720p. #### Push-to-Talk Right Option = dictate anywhere (STT to paste). Right Command = voice command (STT to LLM with tool definitions to ActionRouter). System-wide hotkeys. #### Web Scraping CDP-based scraper using a real browser. 16-framework detection, readiness-aware extraction, 7 output formats: markdown, text, readability, AX tree, HTML, links, screenshot. ## FAQ ### Do I need Ollama? No. ToolPiper embeds the same llama.cpp engine Ollama wraps, directly rather than as a fork, and the whole runner is free: unlimited GGUF downloads, multi-model switching, the local OpenAI-compatible API, no account. Models are stored as standard GGUF files you can use with any llama.cpp tool, not a proprietary blob format. If you keep Ollama installed, ModelPiper can still connect to it as a backend, but there's nothing it provides that ToolPiper doesn't run natively. ### Will my Ollama tools keep working? Yes - ToolPiper serves the Ollama API itself. Flip on the compatibility listener (Settings → General, off by default) and anything that talks to localhost:11434 talks to ToolPiper: model list, streamed chat, embeddings, pulls, deletes, all on the native engine. It's served as the legacy dialect - every response carries a standards-based deprecation header pointing at the first-party /v1 API, which is where new integrations should land. ### What models can I run? Any GGUF model from HuggingFace (135,000+ available). ToolPiper also ships curated presets tested on Apple Silicon: Qwen 3.5, Llama 3.2, DeepSeek R1, Phi-4, Gemma 3, and more. Each preset shows exact RAM usage so you never download something your Mac can't run. ### Does it work with Claude Code? Yes. One command: claude mcp add toolpiper -- ~/.toolpiper/mcp. Restart Claude Code and all over 420 tools are available. Also works with Cursor, Windsurf, and any MCP-compatible client via stdio or Streamable HTTP transport. ### How much RAM do I need? 8GB minimum for small models (0.8B-3B). 16GB recommended for the mainstream sweet spot (7B-8B models alongside normal app usage). 32GB opens up 14B models and multi-model workflows. ToolPiper's resource intelligence monitors RAM in real-time and automatically evicts models under memory pressure. ### Is it really free? Yes, and free is most of the product. The native llama.cpp engine, unlimited model downloads, multi-model switching, the local OpenAI-compatible API, embeddings, developer tokens, all speech (transcription, text-to-speech, voice cloning, push-to-talk dictation, voice chat), vision and OCR, Apple Intelligence, full browser automation, system control, and the visual pipeline builder are free with no account and no caps: 358 of ToolPiper's 421 tools. Pro ($10/mo) adds local RAG over your files, web scraping with YouTube transcripts, and the cloud API proxy for your own keys. Studio ($29/mo) adds image and video upscaling, video editing, pose detection, and the outreach toolkit. Max ($49/mo) adds CodePiper, PiperTest with self-healing, API discovery and replay, and priority support. ### Does it work offline? Yes, completely. All inference runs on your Mac's hardware. Once models are downloaded, everything works without an internet connection: chat, transcribe, speak, embed, OCR, browser automation, testing, and desktop control. The only features that require network access are cloud API proxy and the social/research tools (GitHub, Hacker News, Reddit). URL: https://modelpiper.com/visionpiper # VisionPiper: Screen Capture with Live AI Streaming for Mac > Capture any screen region, record video, export GIFs, and stream live to vision models. A steerable camera for your screen. Free on the Mac App Store. **Free on Mac App Store** ## Screen capture that talks to AI. Capture any region, record video, export GIFs, and stream live to vision models. A steerable camera for your screen. VisionPiper is a standalone menu bar app for macOS that captures screen regions with precision borders, records H.264 video, trims and exports as GIF/WebP/MP4, and streams frames over WebSocket to ToolPiper's vision models at 30fps. Think of it as a screen-mounted camera you can point at anything. - [Download from Mac App Store](https://apps.apple.com/us/app/visionpiper/id6759798951?mt=12) - [See vision AI workflows](/workflows/vision-ai) ## Precision Screen Capture #### 8-Window Border Four edges + four corners, pass-through interior. Click through the capture region to interact with everything inside it normally. #### Movable While Recording Drag the border to follow content while recording is active. A steerable camera that tracks multi-step workflows across panels. #### Region Persistence Last capture region is remembered between sessions. Launch VisionPiper and your previous region is already set. #### Keyboard Shortcuts Quick capture, toggle recording, adjust region from keyboard. Fast enough for rapid bug documentation. #### Multi-Monitor Works across all connected displays. Place the capture region on any monitor regardless of resolution or scaling. #### High DPI Retina-aware capture at native resolution. Every pixel captured at the display's actual density. ## Recording & Export #### H.264 Video SCStream + AVAssetWriter for efficient, hardware-accelerated recording. Captures the selected region as .mp4 with minimal CPU overhead. #### GIF Export Built-in cgif + libimagequant for high-quality, small-file GIFs. Allocates color information to changing pixels, not static backgrounds. #### WebP Export Modern format for web-ready screen captures. Full color and alpha channel support, smaller files than GIF. #### Trim Editor Cut start and end of recordings before export. No external editor needed. Preview, trim, and export in one step. ## Live AI Streaming #### 30fps WebSocket Stream Metal JPEG frames streamed to ToolPiper on port 10000. Hardware-accelerated encoding keeps CPU usage low during continuous streaming. #### Screen Q&A Ask AI questions about what's on your screen in real time. The vision model sees exactly what you see, updated every frame. #### Image Narration Vision model describes the screen content, TTS reads it aloud. An accessibility workflow that works with any on-screen content. #### OCR Pipeline Extract text from any screen region using Apple Vision OCR. No cloud API, no rate limits. Runs entirely on-device. ## FAQ ### Does it work without ToolPiper? Yes. Screen capture, video recording, trim editing, and GIF/WebP/MP4 export all work standalone. ToolPiper is only needed for AI streaming, sending frames to vision models for Screen Q&A, image narration, and OCR. ### What macOS version is required? macOS 26 or later on Apple Silicon (M1 or newer). VisionPiper uses ScreenCaptureKit and Metal, which require Apple Silicon. ### Can I click through the capture region? Yes. The 8-window architecture uses four edge windows and four corner windows with a pass-through interior. Everything inside the border is fully interactive: clicks, drags, and scrolls pass through to the underlying app. ### Does it record system audio? No. VisionPiper captures video only. For audio capture, use AudioPiper, a separate free app that records mic, system audio, and per-app audio via Core Audio Taps. ## Workflows > 9 entries below. Each row is a one-line summary; fetch the linked `.md` for the full body. - **AI Testing on Mac: Visual, Self-Healing, Accessibility-Native Browser Testing** (2026-03-31) — https://modelpiper.com/workflow/ai-testing The definitive guide to AI-powered browser testing. PiperTest uses Chrome's real accessibility tree, self-heals broken selectors, and exports to Playwright or Cypress. - **AI Upscale on Mac: Image and Video Super Resolution on Apple Neural Engine** (2026-03-22) — https://modelpiper.com/workflow/ai-upscale The definitive guide to AI super resolution on Mac. PiperSR upscales images 2-4x and video at 44 FPS on Apple Neural Engine. No cloud. Studio ($29/mo), no per-image fees. - **AI Developer Tools on Mac: MCP, Browser Automation, and Local APIs** (2026-03-27) — https://modelpiper.com/workflow/developer-platform The definitive guide to local AI developer infrastructure on Mac. MCP server, OpenAI-compatible API, browser automation, AI agents, and tool calling. - **Document AI on Mac: OCR, RAG, Embeddings, and Code Search Without the Cloud** (2026-03-21) — https://modelpiper.com/workflow/document-ai The definitive guide to document AI workflows on Mac. RAG pipelines, OCR, embeddings, code search, and web scraping — all local on Apple Silicon. - **Local AI Chat on Mac: Models, Tools, and the State of Private Inference** (2026-03-15) — https://modelpiper.com/workflow/local-chat The definitive guide to running AI chat locally on Mac. Models, hardware requirements, speed benchmarks, and how ToolPiper makes private inference effortless. - **AI Automation on Mac: Voice Commands, Desktop Control, and System Actions** (2026-03-30) — https://modelpiper.com/workflow/mac-automation The definitive guide to AI-powered Mac automation. Voice commands, 26 action domains, MCP tools, and push-to-talk - all local, no scripting required. - **Text Expansion and Clipboard on Mac: Snippets, History, and AI Transforms** (2026-05-15) — https://modelpiper.com/workflow/text-clipboard The reference guide to text expansion and clipboard on Mac. Snippets, clipboard history, dictation, and AI transforms in one local system. Free download. - **Vision AI on Mac: Screen Analysis, Image Understanding, and Pose Detection** (2026-03-20) — https://modelpiper.com/workflow/vision-ai The definitive guide to vision AI workflows on Mac. Run multimodal LLMs, analyze screen content, narrate images, and track human poses - all locally on Apple Silicon. Image Q&A, screen analysis, and OCR are free; pose detection is Studio. - **Voice AI on Mac: Transcription, Speech, and Conversation — All Local** (2026-03-16) — https://modelpiper.com/workflow/voice-ai The definitive guide to local voice AI on Mac. Speech-to-text, text-to-speech, voice cloning, live transcription, and voice commands - all on Apple Silicon. ## Pages > 132 entries below. Each row is a one-line summary; fetch the linked `.md` for the full body. - **ModelPiper: Local AI, Private by Default. Talk to Your Mac.** — https://modelpiper.com/ Private, local AI for Mac. Talk to it to run LLMs, build visual pipelines, and automate macOS on-device. Supports Ollama, OpenAI, Anthropic, Groq. No cloud. - **About ModelPiper & Founder Ben Racicot** — https://modelpiper.com/about ModelPiper is a one-person studio building local-first AI tools for macOS. Ben Racicot designs and ships every layer: Swift apps, backends, web app, MCP server. - **Blog | ModelPiper** — https://modelpiper.com/blog Technical papers, product updates, and press releases from the ModelPiper team. Local AI, MCP servers, CoreML optimization, and more. - **Changelog | ModelPiper** — https://modelpiper.com/changelog Versioned release notes for ModelPiper, ToolPiper, VisionPiper, AudioPiper, and MediaPiper. Newest first. - **ToolPiper Compared: Ollama, LM Studio, Msty, Jan, Wispr Flow, and More** — https://modelpiper.com/compare Every ToolPiper comparison in one place. Local AI platforms, chat apps, dictation, Mac utilities, and developer tools, side by side with honest trade-offs. - **Claude Code Backends Compared: ToolPiper vs Ollama vs vLLM vs LM Studio** — https://modelpiper.com/compare/claude-code-backends Side-by-side comparison of every Anthropic-compatible local backend for Claude Code. Features, performance, BYOK support, and where each one wins. - **Is LM Studio Free? The Local App and the New Cloud Tier** — https://modelpiper.com/compare/is-lm-studio-free LM Studio's desktop app is free. Its cloud inference is billed per million tokens. Here is the boundary, checked against their pricing page. - **Is Ollama Free? What Costs Money and What Does Not** — https://modelpiper.com/compare/is-ollama-free Running Ollama locally is free. Ollama Cloud is a paid product with published tiers. Here is the split, checked against their pricing page. - **Dictation Software for Mac: On-Device, 127 ms, Free** — https://modelpiper.com/dictation-mac ToolPiper dictation transcribes speech in 127-173 ms on the Apple Neural Engine. Hold the right Option key and speak. No cloud, no account, no subscription. - **Documentation | ModelPiper** — https://modelpiper.com/docs Documentation for ModelPiper, ToolPiper, VisionPiper, AudioPiper, and MediaPiper. API references, guides, and integration docs. - **AudioPiper Docs | ModelPiper** — https://modelpiper.com/docs/audiopiper Documentation for AudioPiper. Speech-to-text, text-to-speech, and audio processing pipelines running locally. - **MCP Integration | ModelPiper** — https://modelpiper.com/docs/mcp Model Context Protocol (MCP) server documentation for ToolPiper. Connect Claude Code, AI agents, and other MCP clients to local AI models. - **MediaPiper Docs | ModelPiper** — https://modelpiper.com/docs/mediapiper MediaPiper browser extension: hover any image or video for full-size preview, with on-device AI upscaling and 40+ shortcuts. Chrome, Firefox, Safari. - **ModelPiper Web App Docs | ModelPiper** — https://modelpiper.com/docs/modelpiper Documentation for the ModelPiper visual pipeline builder web application. Block types, pipeline concepts, and workflow guide. - **ToolPiper API Reference | ModelPiper** — https://modelpiper.com/docs/toolpiper API documentation for ToolPiper, the macOS menu bar app for local AI model inference. REST API endpoints, model management, and MCP server integration. - **VisionPiper Docs | ModelPiper** — https://modelpiper.com/docs/visionpiper Documentation for VisionPiper. Menu bar screen capture and recording for macOS: region select, MP4 recording, GIF and WebP export, on-device text extraction. - **Download ToolPiper: Local AI Engine for Mac** — https://modelpiper.com/download Download ToolPiper for macOS. On-device inference, browser automation, RAG, voice, and over 420 MCP tools. Apple Silicon, macOS 26+. Free to download. - **Features | ModelPiper** — https://modelpiper.com/features The full ModelPiper feature portfolio: apps, inference, image and video, developer platform, productivity, and networking, with status for every capability. - **Can I Run This LLM? Mac AI Model Checker | ModelPiper** — https://modelpiper.com/fit Check which AI models your Mac can run locally. Apple Silicon optimized for M1, M2, M3, M4. See fit ratings, estimated speed, and recommended quantizations. - **iMac for Local AI: What Each Chip Can Run | ModelPiper** — https://modelpiper.com/imac Which LLMs run on an iMac. Base M chips only, so unified memory is the limit that matters. Every generation with bandwidth, memory options, and model fit ratings. - **iMac M1: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/imac/m1 Local AI on the M1 iMac: 68.25 GB/s memory bandwidth and 8 to 16 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **iMac M3: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/imac/m3 Local AI on the M3 iMac: 100 GB/s memory bandwidth and 8 to 24 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **iMac M4: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/imac/m4 Local AI on the M4 iMac: 120 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Try ModelPiper: Local, Private AI for Mac** — https://modelpiper.com/login Sign in or create a free ModelPiper account. Visual AI pipeline builder for Mac. Run LLMs, vision, and TTS locally. No cloud required. - **Mac mini for Local AI: What Each Chip Can Run | ModelPiper** — https://modelpiper.com/mac-mini Which LLMs run on a Mac mini, the cheapest way into local AI on Apple Silicon. Every generation with memory bandwidth, unified memory options, and model fit ratings. - **Mac mini M1: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-mini/m1 Local AI on the M1 Mac mini: 68.25 GB/s memory bandwidth and 8 to 16 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Mac mini M2 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-mini/m2-pro Local AI on the M2 Pro Mac mini: 200 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Mac mini M2: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-mini/m2 Local AI on the M2 Mac mini: 100 GB/s memory bandwidth and 8 to 24 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Mac mini M4 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-mini/m4-pro Local AI on the M4 Pro Mac mini: 273 GB/s memory bandwidth and 24 to 64 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Mac mini M4: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-mini/m4 Local AI on the M4 Mac mini: 120 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Mac Studio for Local AI: What Each Chip Can Run | ModelPiper** — https://modelpiper.com/mac-studio Which LLMs run on a Mac Studio. Max and Ultra chips carry the highest memory bandwidth and unified memory ceilings Apple ships, up to 512 GB. Fit ratings for every model. - **Mac Studio M1 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-studio/m1-max Local AI on the M1 Max Mac Studio: 400 GB/s memory bandwidth and 32 to 64 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Mac Studio M1 Ultra: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-studio/m1-ultra Local AI on the M1 Ultra Mac Studio: 800 GB/s memory bandwidth and 64 to 128 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Mac Studio M2 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-studio/m2-max Local AI on the M2 Max Mac Studio: 400 GB/s memory bandwidth and 32 to 96 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Mac Studio M2 Ultra: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-studio/m2-ultra Local AI on the M2 Ultra Mac Studio: 800 GB/s memory bandwidth and 64 to 192 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Mac Studio M3 Ultra: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-studio/m3-ultra Local AI on the M3 Ultra Mac Studio: 819 GB/s memory bandwidth and 96 to 512 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **Mac Studio M4 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/mac-studio/m4-max Local AI on the M4 Max Mac Studio: 410 to 546 GB/s memory bandwidth and 36 to 128 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Air for Local AI: What Each Chip Can Run | ModelPiper** — https://modelpiper.com/macbook-air Which LLMs run on a MacBook Air. Every M1 through M5 generation with memory bandwidth, unified memory ceilings, and honest fit ratings for local AI on a fanless Mac. - **MacBook Air M1: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-air/m1 Local AI on the M1 MacBook Air: 68.25 GB/s memory bandwidth and 8 to 16 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Air M2: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-air/m2 Local AI on the M2 MacBook Air: 100 GB/s memory bandwidth and 8 to 24 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Air M3: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-air/m3 Local AI on the M3 MacBook Air: 100 GB/s memory bandwidth and 8 to 24 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Air M4: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-air/m4 Local AI on the M4 MacBook Air: 120 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Air M5: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-air/m5 Local AI on the M5 MacBook Air: 153 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" for Local AI: What Each Chip Can Run | ModelPiper** — https://modelpiper.com/macbook-pro-14 Which LLMs run on a 14-inch MacBook Pro. Base, Pro, and Max chips compared by memory bandwidth and unified memory, with model fit ratings and speed estimates. - **MacBook Pro 14" M1 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m1-max Local AI on the M1 Max MacBook Pro 14": 400 GB/s memory bandwidth and 32 to 64 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M1 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m1-pro Local AI on the M1 Pro MacBook Pro 14": 200 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M2 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m2-max Local AI on the M2 Max MacBook Pro 14": 400 GB/s memory bandwidth and 32 to 96 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M2 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m2-pro Local AI on the M2 Pro MacBook Pro 14": 200 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M3 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m3-max Local AI on the M3 Max MacBook Pro 14": 300 to 400 GB/s memory bandwidth and 36 to 128 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M3 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m3-pro Local AI on the M3 Pro MacBook Pro 14": 150 GB/s memory bandwidth and 18 to 36 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M3: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m3 Local AI on the M3 MacBook Pro 14": 100 GB/s memory bandwidth and 8 to 24 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M4 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m4-max Local AI on the M4 Max MacBook Pro 14": 410 to 546 GB/s memory bandwidth and 36 to 128 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M4 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m4-pro Local AI on the M4 Pro MacBook Pro 14": 273 GB/s memory bandwidth and 24 to 48 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M4: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m4 Local AI on the M4 MacBook Pro 14": 120 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M5 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m5-max Local AI on the M5 Max MacBook Pro 14": 460 to 614 GB/s memory bandwidth and 36 to 128 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M5 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m5-pro Local AI on the M5 Pro MacBook Pro 14": 307 GB/s memory bandwidth and 24 to 64 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 14" M5: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-14/m5 Local AI on the M5 MacBook Pro 14": 153 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 16" for Local AI: What Each Chip Can Run | ModelPiper** — https://modelpiper.com/macbook-pro-16 Which LLMs run on a 16-inch MacBook Pro. Pro and Max chips from M1 to M5, ranked by memory bandwidth and unified memory, with model fit ratings and speed estimates. - **MacBook Pro 16" M1 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-16/m1-max Local AI on the M1 Max MacBook Pro 16": 400 GB/s memory bandwidth and 32 to 64 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 16" M1 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-16/m1-pro Local AI on the M1 Pro MacBook Pro 16": 200 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 16" M2 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-16/m2-max Local AI on the M2 Max MacBook Pro 16": 400 GB/s memory bandwidth and 32 to 96 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 16" M2 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-16/m2-pro Local AI on the M2 Pro MacBook Pro 16": 200 GB/s memory bandwidth and 16 to 32 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 16" M3 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-16/m3-max Local AI on the M3 Max MacBook Pro 16": 300 to 400 GB/s memory bandwidth and 36 to 128 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 16" M3 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-16/m3-pro Local AI on the M3 Pro MacBook Pro 16": 150 GB/s memory bandwidth and 18 to 36 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 16" M4 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-16/m4-max Local AI on the M4 Max MacBook Pro 16": 410 to 546 GB/s memory bandwidth and 36 to 128 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 16" M4 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-16/m4-pro Local AI on the M4 Pro MacBook Pro 16": 273 GB/s memory bandwidth and 24 to 48 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 16" M5 Max: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-16/m5-max Local AI on the M5 Max MacBook Pro 16": 460 to 614 GB/s memory bandwidth and 36 to 128 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MacBook Pro 16" M5 Pro: What AI Models Can It Run? | ModelPiper** — https://modelpiper.com/macbook-pro-16/m5-pro Local AI on the M5 Pro MacBook Pro 16": 307 GB/s memory bandwidth and 24 to 64 GB of unified memory. See which LLMs fit, at which quantization, and how fast they run. - **MCP Tools for ToolPiper: Curated Marketplace** — https://modelpiper.com/mcp-tools Browse curated MCP servers for ToolPiper. Install GitHub, Slack, Postgres, and more with one click. Tools work with Claude, ChatGPT, and local models. - **arXiv MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/arxiv Search, download, and read arXiv research papers. The arXiv MCP server lets the assistant search the arXiv preprint corpus, download papers by ID, and read their full text. Downloaded PDFs are cached in a folder you choo - **AWS Documentation MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/aws-docs Search and read official AWS documentation. AWS Labs maintains a family of official MCP servers; this one puts the entire AWS documentation set at the assistant's fingertips. Search across services, read any doc page as - **AWS Knowledge Base MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/aws-kb-retrieval Retrieve answers from an AWS Bedrock Knowledge Base. One tool: `retrieve_from_aws_kb`. Takes a `query` string, a `knowledgeBaseId`, and an optional `n` (default 3) controlling how many passages come back. The server call - **Azure MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/azure Query and manage Azure resources through the Azure CLI auth. Microsoft's Azure MCP server lets the assistant explore and operate your Azure estate using the credentials you already have from az login. List subscriptions - **Brave Search MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/brave-search Web and local search via Brave's privacy-focused search API. An independently-built search index, not a Bing or Google reseller. The free tier covers 2,000 queries per month at sign-up, and Brave does not log query conte - **Browserbase MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/browserbase Cloud headless browsers the assistant can drive and observe. Browserbase runs real headless browsers in the cloud, so automation does not tie up your machine and pages render exactly as a browser sees them. The MCP serve - **Chroma MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/chroma Vector and full-text search over a Chroma collection. Chroma is an open-source embedding database; this official server exposes a Chroma Cloud collection as a retrieval surface for the assistant. Add documents, list coll - **ClickHouse MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/clickhouse Run analytical SQL against ClickHouse from the assistant. ClickHouse is a columnar database built for fast analytics over large datasets; its official MCP server lets the assistant query it in read-only mode. List databa - **Cloudflare Docs MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/cloudflare-docs Search Cloudflare's developer documentation, no auth required. Cloudflare hosts a family of remote MCP servers; this is the documentation one, and it needs no account or token. The assistant searches Cloudflare's develop - **Context7 MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/context7 Up-to-date, version-specific docs for libraries and frameworks. Context7 fixes the stalest part of coding with an LLM: library docs. It pulls current, version-specific documentation and code examples for thousands of pac - **E2B MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/e2b Run code in secure cloud sandboxes from the assistant. E2B gives the assistant a disposable cloud sandbox to run code in. Instead of the model describing what a script would output, it actually runs it in an isolated con - **Elasticsearch MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/elasticsearch Search and explore Elasticsearch indices from chat. Elastic's official MCP server lets the assistant explore your Elasticsearch data conversationally. List indices, read field mappings, run query DSL or ES|QL searches, a - **ElevenLabs MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/elevenlabs Text-to-speech, voice cloning, and audio transcription. ElevenLabs' MCP server brings high-quality text-to-speech, transcription, and voice cloning into chat. Generate narration in any of your voices, clone a new voice f - **EverArt MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/everart AI image generation using multiple model backends. One tool, `generate_image`, with three parameters: a text `prompt`, a `model` ID, and an optional `image_count`. The server submits the prompt to EverArt's API, EverArt - **Everything MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/everything Reference test server exercising every MCP feature. Everything is not a useful server. It is a reference test fixture that exercises every part of the MCP protocol so client developers can verify their implementation. Pr - **Exa MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/exa Neural web search and content retrieval for AI. Exa is a search engine built for embeddings rather than keywords, so it is good at the queries where Google leans on exact terms: 'find pages like this one' or 'startups do - **Fetch MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/fetch Fetch web pages and return clean markdown. The Fetch MCP server is a lightweight URL fetcher that strips boilerplate and returns markdown your model can actually read. Pair it with a search server for one-stop research. - **Filesystem MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/filesystem Read and write files in a directory you choose. The Filesystem MCP server is the foundational read/write surface for AI assistants. Every MCP user encounters it. Pick a directory; the server scopes all access to that roo - **Firecrawl MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/firecrawl Scrape, crawl, and extract structured data from websites. Firecrawl turns websites into clean markdown or structured JSON. Scrape one page, crawl a whole site, map every URL, or pass a schema and have it extract typed fi - **Git MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/git Read, search, and manipulate Git repositories on disk. Read and write a local working tree from chat. The server shells out to `git` directly, so behavior matches the CLI you already know: status, diff, log, show, branch - **GitHub (Remote) MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/github-remote OAuth-protected GitHub MCP. Manage repos, issues, and pull requests with no token to copy. GitHub's officially hosted MCP server. Same surface as the stdio reference, but the auth path is OAuth instead of a personal acce - **GitHub MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/github Create issues, manage pull requests, and search repositories from your AI assistant. The GitHub MCP server gives your AI assistant first-class access to your repositories. Create issues, open pull requests, and read file - **GitLab MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/gitlab Manage GitLab projects, issues, and merge requests. Nine tools cover the parts of GitLab you actually touch from a code chat: read a file with `get_file_contents`, commit a single file with `create_or_update_file`, batch - **Google Drive MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/google-drive Search and read files from Google Drive. Read-only by design. Two affordances: a `search` tool that takes a query string and returns matching files (name plus MIME type), and a `gdrive:///` resource pattern the - **Google Maps MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/google-maps Geocoding, directions, and place lookup via Google Maps. Geocoding, reverse geocoding, place search, place details, distance matrix, elevation, and directions. Seven tools that cover what a chat agent typically reaches f - **Grafana MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/grafana Query dashboards, datasources, and alerts from Grafana. Grafana Labs' MCP server lets the assistant reach into your observability stack: find dashboards, read their panels, query Prometheus and Loki datasources, and insp - **Heroku MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/heroku Manage Heroku apps, dynos, add-ons, and config from chat. Heroku's official MCP server lets the assistant operate your apps through the Platform API: list apps, read config and add-ons, inspect and scale dynos, run one-o - **Kubernetes MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/kubernetes Inspect and manage Kubernetes clusters via your kubeconfig. This server gives the assistant the read and write surface of kubectl, scoped to whatever your current kubeconfig context allows. List pods, read logs, get and - **Memory MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/memory Knowledge-graph backed persistent memory across chats. Memory keeps a JSONL knowledge graph on disk and gives the model nine tools to read and write it. Entities have a name, a type (`person`, `project`, `event`), and a - **MongoDB MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/mongodb Query and manage MongoDB and Atlas clusters from the assistant. MongoDB's official MCP server connects to a cluster directly or to Atlas via the Admin API. With a connection string it can run find and aggregate queries, - **MotherDuck MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/motherduck Run DuckDB and MotherDuck SQL analytics from the assistant. DuckDB is a fast analytical SQL engine; MotherDuck is its cloud. This server points the assistant at either one. With md: as the database path and a MotherDuck - **Neon MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/neon Manage Neon serverless Postgres branches and run SQL. Neon is serverless Postgres with instant branching; this official server lets the assistant use that branching as a safety net. List projects, run SQL, describe schem - **Notion MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/notion Search, read, and update Notion pages and databases from chat. Notion's own MCP server gives your assistant scoped access to the pages and databases you share with an internal integration. Search the workspace, read a pa - **PagerDuty MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/pagerduty Read incidents, services, and on-call schedules from PagerDuty. PagerDuty's MCP server lets the assistant answer the questions you ask during an incident: what is firing, who is on call, which service owns it, and what t - **PayPal MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/paypal Create invoices, orders, and payments through the PayPal API. PayPal's MCP server wraps the PayPal REST API so the assistant can create invoices, orders, and payments and read transaction history. Set the environment to - **Perplexity MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/perplexity Ask Perplexity's Sonar models for cited, up-to-date answers. Perplexity's MCP server lets the assistant hand a question to a Sonar model and get back an answer grounded in live web sources with citations. Use it when the - **Playwright MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/playwright Drive a real browser via accessibility-tree automation. Microsoft's Playwright MCP server automates a real Chromium, Firefox, or WebKit browser using the page's accessibility tree rather than brittle CSS selectors. The m - **Postgres MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/postgres Query Postgres databases with read-only SQL. Point the Postgres MCP server at any Postgres database and let your AI assistant write the SELECTs for you. The server is read-only by construction; DDL and DML are refused. - **Puppeteer MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/puppeteer Headless browser automation and web scraping. Give the assistant a real Chromium to drive. Seven tools map to standard Puppeteer operations: `puppeteer_navigate`, `puppeteer_screenshot`, `puppeteer_click`, `puppeteer_hov - **Qdrant MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/qdrant Store and search vector embeddings as a semantic memory. Qdrant is a vector database; this server turns it into a semantic memory the assistant can write to and search. Store snippets of text with metadata, then retrieve - **Redis MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/redis Interact with Redis key-value stores from the assistant. Four tools: `set` with optional TTL via `expireSeconds`, `get`, `delete` (key or array of keys), and `list` (glob pattern, default `*`). That is the whole surface. - **Sentry MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/sentry Retrieve and analyze issues from Sentry.io. You see a Sentry link in Slack or PagerDuty, paste the issue ID into chat, and the model pulls the trace and starts on root cause. That is the workflow this server is built for - **Sequential Thinking MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/sequential-thinking Dynamic, reflective problem-solving through thought sequences. Force the model into multi-step reasoning instead of a one-shot answer. The single tool, `sequential_thinking`, gets called repeatedly while the model works - **Slack MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/slack Post messages, read channels, and search history in Slack. Connect your AI assistant to Slack so it can read context from channels and post on your behalf. Use it to triage support requests, summarize threads, or draft r - **SQLite MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/sqlite Query SQLite databases and run business-intelligence prompts. Point this at a `.db` file and the model can run SQL against it, read and write. Six tools: `read_query` (SELECT), `write_query` (INSERT/UPDATE/DELETE), `crea - **Stripe MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/stripe Create payments, customers, and invoices through the Stripe API. Stripe's agent-toolkit MCP server wraps the Stripe API so the assistant can create customers, products, prices, payment links, and invoices, and read balan - **Supabase MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/supabase Manage Supabase projects, run SQL, and inspect schemas. Supabase's official server gives the assistant access to your Postgres-backed projects: list projects, run SQL, inspect tables and schemas, read logs, and (in write - **Tavily MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/tavily Web search and extraction tuned for AI agents. Tavily is a search API built for agents: results come back ranked and ready to feed a model, and the extract tool pulls clean content from URLs without the boilerplate. Beyo - **Terraform MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/terraform Look up Terraform providers, modules, and registry docs. HashiCorp's Terraform MCP server gives the assistant accurate, current information from the Terraform Registry: provider resource docs, module inputs and outputs, - **Time MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/time Time and timezone conversion utilities. LLMs are unreliable narrators about 'now'. Training cutoff is months or years stale, prompt timestamps are often missing or wrong, and DST math is a known weak spot. A small Python - **VisionPiper MCP Server for ToolPiper · ModelPiper** — https://modelpiper.com/mcp-tools/visionpiper Drive ToolPiper's screen-vision companion: steer the capture region, grab screenshots, record and export - all on-device. VisionPiper is the ModelPiper companion app for continuous screen work. Federating it into ToolPip - **ToolPiper MCP Server — over 420 Tools, Install Only What You Need** — https://modelpiper.com/mcp Local MCP server for macOS with over 420 tools for AI coding assistants — LLM inference, browser automation, testing, system actions, video, and more. Filter by category to save context window tokens. Works with Claude Code, Cursor, Windsurf. - **Every ToolPiper MCP Tool — Searchable Catalog of over 420 Tools** — https://modelpiper.com/mcp/tools The full catalog of built-in ToolPiper MCP tools, searchable and organized by category with the unlock tier for every tool. Inference, browser, macOS, video, RAG. - **Press & Media | ModelPiper** — https://modelpiper.com/press ModelPiper press kit. Logo downloads, copy-paste boilerplate, published press releases, and a media contact for press coverage. - **Pricing | ModelPiper** — https://modelpiper.com/pricing ModelPiper plans: Free, Pro, Studio, Max. Local-first AI on your Mac. Subscribe monthly or annually. - **Privacy Policy | ModelPiper** — https://modelpiper.com/privacy Privacy Policy for ModelPiper. All AI processing runs locally on your device. Your prompts, responses, and workflows never leave your machine. - **Recipes: Point Claude Code at a Model You Control | ModelPiper** — https://modelpiper.com/recipes Complete setups for running Claude Code on Apple Intelligence, a local Qwen model, or your own OpenAI key, and for switching providers mid-session. - **Run Claude Code on Apple Intelligence: Free, On-Device, No API Key** — https://modelpiper.com/recipes/claude-code-apple-intelligence Point Claude Code at Apple Intelligence on your Mac's Neural Engine. Free, private, no cloud round-trip, no per-token billing. Setup in two minutes. - **Use Claude Code with Your OpenAI Key: Keychain-Locked BYOK** — https://modelpiper.com/recipes/claude-code-byok-openai Bring your own OpenAI key to Claude Code. ToolPiper injects it server-side from macOS Keychain. Never in dotfiles, never in logs. - **Switch Claude Code Providers In Chat: "Use My Local One"** — https://modelpiper.com/recipes/claude-code-cross-provider Switch backends from inside Claude Code with natural language. ToolPiper's endpoint_set MCP tool reroutes the session live. No env vars, no restart. - **Run Claude Code on Local Qwen: 32K Context, Tool Use, Fully Private** — https://modelpiper.com/recipes/claude-code-local-qwen Wire Claude Code to a local Qwen model on your Mac via llama.cpp. Best balance of capability and privacy: long context, tool use, no API key. - **Terms of Service | ModelPiper** — https://modelpiper.com/terms Terms of Service for ModelPiper, ToolPiper, and VisionPiper products. - **AI Workflows on Mac | ModelPiper** — https://modelpiper.com/workflow Definitive guides to local AI workflows on Mac. Models, tools, comparisons, and step-by-step setup for chat, voice, vision, document AI, and more. ## Articles > 153 entries below. Each row is a one-line summary; fetch the linked `.md` for the full body. - **Ollama MCP Server on Mac: Three Bridges, One Native Server** (2026-08-12) — https://modelpiper.com/blog/ollama-mcp-server-mac Ollama ships no MCP server. What the bridge projects do, which direction each one runs, and what a native macOS MCP server does instead. - **LM Studio Alternative: The Honest List (2026)** (2026-08-10) — https://modelpiper.com/blog/lm-studio-alternative Every real LM Studio alternative in 2026, sorted by what you are actually replacing: the engine, the model browser, the chat window, or the whole app. - **LM Studio Failed to Load Model on Mac: The Four Causes** (2026-08-10) — https://modelpiper.com/blog/lm-studio-failed-to-load-model-mac A model that will not load on Apple Silicon has one of four causes. Here is how to tell them apart in under a minute, and the fix for each. - **LM Studio GPU Offload on Mac: What the Slider Actually Does** (2026-08-10) — https://modelpiper.com/blog/lm-studio-gpu-offload-mac On Apple Silicon there is no graphics card to offload to. Here is what LM Studio's GPU offload setting really controls, and the memory ceiling you are hitting. - **LM Studio Not Working on Mac: Won't Open, Won't Respond, Not Supported** (2026-08-10) — https://modelpiper.com/blog/lm-studio-not-working-mac Three different failures wear the same phrase. Check the requirements first, because one of the three is not fixable and the other two take a minute each. - **LM Studio Slow on Mac: Prompt Processing Is Usually the Culprit** (2026-08-10) — https://modelpiper.com/blog/lm-studio-slow-mac Slow local inference on Apple Silicon is usually prompt processing, not generation. The difference decides which fix works, and most advice targets the wrong one. - **Migrate from LM Studio on Mac: Your Models Are Already Portable** (2026-08-10) — https://modelpiper.com/blog/migrate-from-lm-studio-mac There is no export step. LM Studio stores plain GGUF files, so moving to another tool is a file move. Here is the whole migration and what stays behind. - **Migrate from Open WebUI on Mac: Get Your Data Out of the Volume** (2026-08-10) — https://modelpiper.com/blog/migrate-from-open-webui-mac Your models were never inside Open WebUI. Your conversations were. Here is how to get them off a Docker volume on macOS before you tear the container down. - **Uninstall LM Studio on Mac: The Model Cache Does Not Go With the App** (2026-08-10) — https://modelpiper.com/blog/uninstall-lm-studio-mac Dragging LM Studio to the Trash leaves your models behind, often tens of gigabytes. Here is every location to check, and which one you may want to keep. - **ToolPiper Beta 9: Keys in the macOS Keychain, Locked to Your Mac** (2026-07-09) — https://modelpiper.com/blog/toolpiper-beta-9-keychain-security ToolPiper beta 9 keeps every token and API key in the macOS Keychain, encrypted and locked to your Mac. One heads-up: you may need to re-add cloud API keys. - **The Danger Is in the Tool, Not the Behavior** (2026-06-25) — https://modelpiper.com/blog/ai-danger-tool-not-behavior When a danger narrative targets a model's capability instead of its misuse, it is arguing to restrict the technology. That choice is never neutral. - **Ollama-Compatible API on Mac: Keep Your Tools, Swap the Server** (2026-06-11) — https://modelpiper.com/blog/ollama-compatible-api-mac ToolPiper serves the Ollama API on localhost:11434. Point the Ollama CLI, ollama-python, the OpenAI SDK, or Hollama at it with no code changes. Native Mac app. - **Replace Ollama on Mac: Everything It Does, Free, Tools Unchanged** (2026-06-11) — https://modelpiper.com/blog/replace-ollama-mac As of beta 7, ToolPiper's free tier covers the whole Ollama surface - and serves the Ollama API itself on localhost:11434, so your tools come with you. - **Why I Built a Suite of AI Tools for macOS** (2026-06-11) — https://modelpiper.com/blog/why-i-built-ai-suite-macos Local AI on Mac was a parts bin, not a product. Why assembling nine inference backends and over 420 MCP tools into one native suite took seven apps. - **Automate Your Mac with AI Agents: Five MCP Servers or One** (2026-06-10) — https://modelpiper.com/blog/automate-macos-ai-agents Wire Claude Code, Cursor, or any MCP client to actually operate your Mac. What five single-purpose servers cost to assemble, and when one app wins. - **Best macOS MCP Servers in 2026: 14 Picks, Verified and Ranked** (2026-06-10) — https://modelpiper.com/blog/best-macos-mcp-servers-2026 The 14 best macOS MCP servers in 2026, verified: ToolPiper, LMCP, Macuse, XcodeBuildMCP, Peekaboo, Apple's mcpbridge, and which popular pick is archived. - **Jan AI Alternative for Mac: Same Architecture, More Platform** (2026-06-10) — https://modelpiper.com/blog/jan-ai-alternative-mac Jan is the genuinely open-source local AI app, we say so up front. If you've hit its walls - no voice, no MCP server, no indexed RAG - here's the alternative. - **What Is a Local AI Engine? The Layer That Runs AI on Your Mac** (2026-06-10) — https://modelpiper.com/blog/local-ai-engine-mac A local AI engine runs models on your own machine and serves them to everything else. What the term means, what qualifies, and where ToolPiper fits. - **Best AI Chat App for Mac? Pick the Engine First** (2026-06-10) — https://modelpiper.com/blog/mac-ai-chat-clients-vs-local-engines Chat clients like BoltAI and Msty are windows onto models that run somewhere else. The engine underneath decides privacy, cost, and offline. Pick it first. - **Msty Alternative for Mac: The Engine Instead of the Manager** (2026-06-10) — https://modelpiper.com/blog/msty-alternative-mac Msty says prices only move in one direction, up. If that sent you looking: the strongest alternative isn't another model manager, it's the engine itself, free. - **Jan vs ToolPiper: Open Source Chat App vs the Mac AI Platform** (2026-06-10) — https://modelpiper.com/blog/toolpiper-vs-jan Jan and ToolPiper agree on the architecture - local llama.cpp, plain GGUF, local API. Jan wins on open source. ToolPiper adds voice, MCP serving, and RAG. - **ToolPiper vs LM Studio: Which Local AI App Does Your Mac Need?** (2026-06-10) — https://modelpiper.com/blog/toolpiper-vs-lm-studio LM Studio vs ToolPiper, compared honestly: engines, MCP server vs client, voice, RAG, pricing, and which privacy posture you can actually verify. - **Msty vs ToolPiper: Same Privacy Promise, Different Layer** (2026-06-10) — https://modelpiper.com/blog/toolpiper-vs-msty Msty and ToolPiper both run AI locally with no-telemetry policies. The real differences: who runs the engine, MCP serving, voice, and what you can verify. - **How to Verify an AI App Is Really Offline on Mac** (2026-06-10) — https://modelpiper.com/blog/verify-ai-app-offline-mac Privacy policies are promises. Sockets are facts. How to check any Mac AI app for outbound calls with Activity Monitor, lsof, nettop, and a free firewall. - **BoltAI Alternative for Mac: The App That Runs the Models** (2026-06-09) — https://modelpiper.com/blog/boltai-alternative-mac BoltAI's license year ends with a renew-or-look-around moment. The honest alternative isn't another chat client - it's the layer that runs the models, free. - **Ollama Export on Mac: Get Your Models Out of the Blob Store** (2026-06-09) — https://modelpiper.com/blog/migrate-from-ollama-mac Ollama has no export command - the request was closed "not planned" in 2026. Here's the manual path from sha256 blobs to plain GGUF files, step by step. - **Ollama's Cloud Pivot: What It Means for Local AI on Mac** (2026-06-09) — https://modelpiper.com/blog/ollama-cloud-privacy Ollama's paid tiers sell cloud inference on partner datacenters. What stays on your Mac, what leaves, what the privacy policy promises, and why direction matters. - **Ollama vs llama.cpp Benchmarks on Apple Silicon (2026)** (2026-06-09) — https://modelpiper.com/blog/ollama-vs-llamacpp-benchmark-mac Same GGUF bytes, same Mac, June 2026: Ollama 0.23.4 vs upstream llama-server b9533. Measured gap: 2-7%, in both directions. Full methodology and commands included. - **BoltAI vs ToolPiper: Mac AI Chat Client vs the Whole Stack** (2026-06-09) — https://modelpiper.com/blog/toolpiper-vs-boltai BoltAI is a polished Mac client for models that run elsewhere. ToolPiper runs them. Where each wins, what each costs, and the one feature BoltAI does better. - **Ollama Environment Variables: The Complete Reference (2026)** (2026-06-02) — https://modelpiper.com/blog/ollama-environment-variables Every Ollama environment variable, audited against the v0.32.5 source: OLLAMA_HOST, OLLAMA_KEEP_ALIVE, OLLAMA_FLASH_ATTENTION, real defaults, and the dead ones. - **Apple Writing Tools Alternative on Mac: Custom AI Text Edits, Local** (2026-05-31) — https://modelpiper.com/blog/apple-writing-tools-alternative An Apple Writing Tools alternative for Mac with custom prompts, chained transforms, your choice of local model, and edits that run on-device. Free download. - **AnythingLLM Alternative for Mac: Bundled Inference, Not Just a Front-End** (2026-05-30) — https://modelpiper.com/blog/anythingllm-alternative-mac AnythingLLM needs a model provider. ToolPiper bundles its own inference and adds RAG, voice, vision, and over 420 MCP tools in one native macOS app. - **LM Studio Alternative for Mac: Same Models, Plus Voice, Vision, and over 420 Tools** (2026-05-30) — https://modelpiper.com/blog/lm-studio-alternative-mac LM Studio runs models with a great browser. ToolPiper runs the same models and adds voice, vision, RAG, browser automation, and an MCP server in one Mac app. - **Local AI Platforms on Mac Compared (2026): Ollama vs LM Studio vs Msty vs Jan vs AnythingLLM vs Open WebUI vs BoltAI vs ToolPiper** (2026-05-30) — https://modelpiper.com/blog/local-ai-platforms-compared-mac Eight local AI tools side by side: which run models, which are front-ends, which privacy claims you can verify, and what each costs. 16-row feature table. - **Open WebUI Alternative for Mac: A Native App Instead of a Docker Stack** (2026-05-30) — https://modelpiper.com/blog/open-webui-alternative-mac Open WebUI is a Docker-based web UI that needs a model server. ToolPiper is a native Mac app bundling inference, voice, vision, RAG, and over 420 MCP tools. - **Espanso Alternative on Mac: Text Expansion Without YAML Config** (2026-05-28) — https://modelpiper.com/blog/espanso-alternative-mac A free, local Espanso alternative for Mac with a visual snippet editor, AI text transforms, dynamic values, clipboard history, and dictation. No YAML to edit. - **Raycast Alternative on Mac (2026): Local AI, Snippets, Clipboard** (2026-05-25) — https://modelpiper.com/blog/raycast-alternative-local-ai Raycast's AI is a paid cloud subscription. ToolPiper runs AI on your text locally, free - snippets, clipboard history, rewrites, plus one-file Raycast import. - **Maccy Alternative on Mac: Clipboard History With AI and OCR** (2026-05-22) — https://modelpiper.com/blog/maccy-alternative-mac A Maccy alternative for Mac with clipboard history, 8-type auto-categorization, image OCR, source-app tracking, and AI text transforms. Free download. - **TextExpander Alternative on Mac (2026): Local, No Subscription** (2026-05-20) — https://modelpiper.com/blog/textexpander-alternative-mac Free TextExpander alternative for Mac: local text expansion, on-device AI rewrites, clipboard history. Imports your .textexpandersettings export, deduped. - **Local Text Expander for Mac: Snippets and Dynamic Values** (2026-05-18) — https://modelpiper.com/blog/text-expander-mac-local A local text expander for Mac: typed-trigger snippets, 11 dynamic values, cursor placement, one-file import from 8 formats. No cloud, no subscription. - **Aider MCP Setup with ToolPiper** (2026-05-17) — https://modelpiper.com/blog/aider-mcp-setup Connect Aider's AI pair-programming CLI to ToolPiper's local MCP server. Over 300 local tools available alongside Aider's repo-aware coding workflow. - **AI Text Snippets on Mac: Rewrite Selected Text With a Keystroke** (2026-05-16) — https://modelpiper.com/blog/ai-text-snippets-mac AI text snippets transform selected text in place - tap right Command to fix grammar, change tone, or summarize. Local model, no cloud, free download. - **Zed MCP Setup with ToolPiper on Mac** (2026-05-16) — https://modelpiper.com/blog/zed-editor-mcp-setup Connect Zed editor to ToolPiper's local MCP server through settings.json. Over 300 local tools available inside Zed's AI assistant. - **VS Code MCP Setup: Add ToolPiper with Cline or Continue on Mac** (2026-05-15) — https://modelpiper.com/blog/vscode-cline-continue-mcp-toolpiper Connect VS Code to ToolPiper's local MCP server through Cline or Continue. Over 300 local tools available inside VS Code agent sessions. - **ToolPiper Browser Extension: Pipe Page Content to Local AI** (2026-05-14) — https://modelpiper.com/blog/toolpiper-browser-extension MediaPiper is the ToolPiper browser extension for Chrome, Firefox, and Safari. Pipe page content to local AI for summary, analysis, RAG, or scraping. - **Use ToolPiper from macOS Shortcuts and Hotkeys** (2026-05-14) — https://modelpiper.com/blog/toolpiper-shortcuts-hotkeys Trigger ToolPiper's local AI from macOS Shortcuts.app and keyboard hotkeys. Push-to-talk dictation, OCR, transcription, system control on demand. - **Connect ToolPiper to Scripts, Shortcuts, and System Triggers on Mac** (2026-05-13) — https://modelpiper.com/blog/connect-toolpiper-scripts-system ToolPiper serves a local HTTP API on 127.0.0.1:9998. Hit it from shell scripts, macOS Shortcuts, hotkeys, or the browser extension. Loopback only. - **ToolPiper HTTP API on Mac: Call over 420 Tools from Any Script** (2026-05-13) — https://modelpiper.com/blog/toolpiper-http-api ToolPiper exposes over 420 tools over a local JSON HTTP API at 127.0.0.1:9998. Shell, Python, Node, anywhere. Loopback needs zero config, tokens off-machine. - **Connect ToolPiper to Your Cloud Accounts via OAuth on Mac** (2026-05-12) — https://modelpiper.com/blog/connect-toolpiper-oauth ToolPiper bundles a Google OAuth client. Connect once, and Search Console tools answer SEO questions from your AI editor. More services rolling out. - **Google Search Console MCP Setup with ToolPiper on Mac** (2026-05-12) — https://modelpiper.com/blog/gsc-search-console-mcp-toolpiper Connect Google Search Console to your AI editor through ToolPiper's MCP server. Four tools: list sites, query analytics, inspect URLs, list sitemaps. - **Claude Desktop MCP Setup on Mac: Add ToolPiper for over 420 Local Tools** (2026-05-11) — https://modelpiper.com/blog/claude-desktop-mcp-setup-mac Connect Claude Desktop to ToolPiper's local MCP server on macOS. JSON config, full app relaunch, over 420 local tools added to every chat. - **Windsurf MCP Setup: Add ToolPiper for over 420 Local Tools on Mac** (2026-05-11) — https://modelpiper.com/blog/windsurf-mcp-setup-toolpiper Connect Windsurf (Codeium) to ToolPiper's local MCP server. JSON config in ~/.codeium/windsurf/, over 420 local tools added to every Cascade session. - **Cursor MCP Setup: Add ToolPiper for over 420 Local Tools on Mac** (2026-05-10) — https://modelpiper.com/blog/cursor-mcp-setup-toolpiper Connect Cursor to ToolPiper's local MCP server with a single JSON snippet. Global or per-project, over 420 local tools, no cloud round-trip. - **Claude Code MCP Setup: Add ToolPiper for over 420 Local Tools on Mac** (2026-05-09) — https://modelpiper.com/blog/claude-code-mcp-setup-toolpiper Connect Claude Code to ToolPiper's local MCP server. One CLI command, over 420 tools, no cloud. Covers user vs project scope, verification, and common gotchas. - **Connect ToolPiper on Mac: AI Editors, Cloud Accounts, and Scripts** (2026-05-08) — https://modelpiper.com/blog/connect-toolpiper ToolPiper exposes over 420 local AI tools to AI editors over MCP, to Google accounts over OAuth, and to scripts over a local HTTP API. - **MCP Setup on Mac: Connect Claude Code, Cursor, Windsurf, and More** (2026-05-08) — https://modelpiper.com/blog/mcp-setup-guide-mac Set up MCP with Claude Code, Cursor, Claude Desktop, Windsurf, VS Code, Zed, or Aider on macOS. One server, over 420 tools, every editor. - **Tool Routing for Local AI on Mac: How Small Models Use over 420 Tools** (2026-05-04) — https://modelpiper.com/blog/local-ai-tool-routing-mac Small local models can't fit a full MCP catalog in 8K context. ToolPiper's on-device retrieval model picks the right tools and shrinks schemas to fit. - **How PiperMesh Works: Direct on LAN, Encrypted Everywhere Else** (2026-05-03) — https://modelpiper.com/blog/how-pipermesh-works PiperMesh tries three paths to connect your devices: direct on LAN, NAT-traversed peer-to-peer, then encrypted relay. All three use Noise encryption end-to-end. - **Conversational AI for Mac: Beyond Dictation, Into Action** (2026-05-02) — https://modelpiper.com/blog/conversational-ai-mac macOS has had dictation since 2012. ToolPiper adds the layer that was always missing: a local LLM that interprets what you said and does something about it. - **Wispr Flow Alternative for Mac: More Features for Less Money** (2026-05-02) — https://modelpiper.com/blog/wispr-flow-alternative-mac ToolPiper's dictation is free against Wispr Flow's $12/month. It runs entirely on your Mac, never sends audio to the cloud, and does far more than dictation. - **Is ToolPiper Safe? What You Can Verify and How** (2026-05-01) — https://modelpiper.com/blog/is-toolpiper-safe ToolPiper's local features process everything on your Mac. No audio, prompts, or documents travel over the network. What stays local, and how to confirm it. - **Token Usage and MCP Tools With ToolPiper** (2026-04-30) — https://modelpiper.com/blog/mcp-tools-token-usage-toolpiper ToolPiper exposes over 420 MCP tools to Claude Code. Three layers keep your context window and token bill in check: deferred loading, permissions, ToolGate. - **ToolPiper vs Wispr Flow: Local Voice AI or Cloud Dictation** (2026-04-30) — https://modelpiper.com/blog/toolpiper-vs-wispr-flow Wispr Flow vs ToolPiper for Mac voice dictation. Wispr Flow is $12/month and processes audio in the cloud. ToolPiper is free and runs on your Neural Engine. - **Wispr Flow's Privacy Incident: What Happened, What Changed, and What It Means** (2026-04-29) — https://modelpiper.com/blog/wispr-flow-privacy-incident In 2025, Wispr Flow users found the app sending audio and screenshots to cloud servers without clear disclosure. What happened, and how the company responded. - **$273 Million in Settlements: The Real Cost of Trusting Cloud Voice AI** (2026-04-28) — https://modelpiper.com/blog/cloud-voice-ai-privacy-settlements Amazon, Apple, and Google have paid over $273 million to settle voice AI privacy claims. The full record: what happened, what each paid, and the pattern. - **Zoom Tried to Train AI on Your Meetings. Here's What Happened.** (2026-04-24) — https://modelpiper.com/blog/zoom-ai-training-controversy In 2023, Zoom granted itself a perpetual license to use meeting content for AI training. The backlash was swift, the terms were revised. What it left behind. - **AI Meeting Recorders and the Consent Problem Nobody Talks About** (2026-04-21) — https://modelpiper.com/blog/ai-meeting-recorder-privacy Otter.ai faces a class action for recording meetings without consent from every participant. What AI meeting tools actually do with your audio, and the risk. - **Amazon's Alexa Listening Program: What Bloomberg Found and What Amazon Paid** (2026-04-21) — https://modelpiper.com/blog/amazon-alexa-listening-program In 2019, Bloomberg revealed Amazon paid thousands of workers to listen to Alexa recordings, including private medical audio. The FTC settlement was $25 million. - **Apple's $95 Million Siri Settlement: What Happened and What It Means** (2026-04-14) — https://modelpiper.com/blog/apple-siri-privacy-settlement Apple settled a class action for $95 million in 2025 over Siri privacy violations. Contractors heard medical conversations, negotiations, and intimate moments. - **Google's $68 Million Voice Privacy Settlement: What the Whistleblower Found** (2026-04-14) — https://modelpiper.com/blog/google-assistant-privacy-settlement Google settled a class action for $68 million in 2026 over Assistant recordings. A whistleblower leaked over 1,000 clips, many captured without a wake word. - **OLLAMA_KV_CACHE_TYPE: Halve Ollama's KV Cache Memory** (2026-04-14) — https://modelpiper.com/blog/ollama-kv-cache-quantization Set OLLAMA_KV_CACHE_TYPE=q8_0 to halve Ollama's KV cache memory, q4_0 to quarter it - near-zero quality loss. Full memory tables, or let ToolPiper default it. - **Ollama vs ToolPiper: The Free Ollama Alternative for Mac** (2026-04-14) — https://modelpiper.com/blog/ollama-vs-toolpiper The free Ollama alternative for Mac: same GGUF models, same llama.cpp engine, a native GUI instead of a terminal - no account, no caps. The full comparison. - **The One-Time License in AI Software Is a Promise Nobody Can Keep** (2026-04-13) — https://modelpiper.com/blog/ai-software-lifetime-license-trap AI tools selling lifetime licenses face a math problem: models change weekly, backends break monthly, revenue per user is zero after day one. Who promises it. - **ChatGPT Alternatives on Mac That Don't Charge Per Token** (2026-04-12) — https://modelpiper.com/blog/chatgpt-alternatives-no-per-token-fees Six ChatGPT alternatives compared by what you actually pay for: token-metered subscriptions, bring-your-own-key apps, and local inference on Apple Silicon. - **Run Multiple Ollama Models on Mac at Once: What Fits in 8, 16, 32GB** (2026-04-10) — https://modelpiper.com/blog/ollama-multi-model-mac How many Ollama models fit in memory at once? 8GB runs one, 16GB two small models, 32GB several. ToolPiper shows live per-model RAM and warns before swapping. - **Ollama Vision GUI on Mac 2026: Drag an Image Into LLaVA or Gemma 3** (2026-04-10) — https://modelpiper.com/blog/ollama-vision-gui-mac Drag an image into chat and ask about it. ToolPiper runs LLaVA, Gemma 3 and Llama 3.2 Vision natively on Mac in 2026, then chains vision into OCR and speech. - **Give Claude over 420 macOS Superpowers with One MCP Server** (2026-04-09) — https://modelpiper.com/blog/mcp-tools-claude-macos-toolpiper Install ToolPiper and connect over 420 MCP tools to Claude Code in under a minute. Local inference, browser automation, desktop control, testing, and video. - **Tracing Multi-Step AI Pipelines with Correlation IDs** (2026-04-07) — https://modelpiper.com/blog/correlation-ids-ai-pipeline-tracing When a 3-step AI pipeline returns wrong output, which step failed? Correlation IDs group every event from one execution into a single trace. - **Full HTTP Body Capture: See Exactly What Your Local AI Sent and Received** (2026-04-07) — https://modelpiper.com/blog/http-body-capture-local-ai-debugging LogPiper captures full request and response bodies for every API call through ToolPiper. When your local LLM returns garbage, the exact prompt is in the logs. - **Ollama Pipelines on Mac: Chain Models in a Visual Workflow** (2026-04-07) — https://modelpiper.com/blog/ollama-pipelines-mac Ollama runs one model at a time. ToolPiper's pipeline builder chains several models into workflows on its built-in llama.cpp engine, or your Ollama models. - **Ollama TTS and STT on Mac: Add Voice Models to Any Local LLM** (2026-04-07) — https://modelpiper.com/blog/ollama-voice-chat-mac Ollama ships no speech models of its own. Chain Parakeet speech-to-text and three local TTS voices onto any Ollama model on your Mac, entirely on-device. - **The Samsung ChatGPT Leak: What Developers Need to Know** (2026-04-07) — https://modelpiper.com/blog/samsung-chatgpt-code-leak In April 2023, Samsung engineers pasted proprietary chip source code into ChatGPT three times in 20 days. Samsung banned AI tools company-wide within a month. - **Why Voice AI Should Never Leave Your Device** (2026-04-07) — https://modelpiper.com/blog/why-voice-ai-should-stay-local Cloud voice AI sends your most personal data to servers you cannot audit, with a documented history of human reviewers listening in. Why local inference wins. - **Cross-App Debugging on macOS Without Installing a Logging Framework** (2026-04-06) — https://modelpiper.com/blog/cross-app-debugging-macos-http-logging Python, shell, JavaScript, Swift - all logging to one HTTP endpoint. No SDK, no dependency, no config. LogPiper aggregates logs from any process on localhost. - **Cursor Debug Mode vs. a Persistent Log Bus for AI Debugging** (2026-04-06) — https://modelpiper.com/blog/cursor-debug-mode-vs-persistent-log-bus Cursor Debug Mode gives your AI a temporary log server. LogPiper gives it a persistent, queryable log store. Both solve the same problem differently. - **How to Debug with Claude Code Using a Local Log Bus** (2026-04-06) — https://modelpiper.com/blog/debug-claude-code-mcp-logging Tell Claude Code to POST logs to LogPiper, reproduce the bug, query the errors, hand them back. A persistent debugging feedback loop in four steps. - **Vibe Coding Hits a Wall at 80%. Debugging the Last 20% Without Debugging Skills** (2026-04-06) — https://modelpiper.com/blog/vibe-coding-debugging-last-20-percent AI gets 80% of the code right. The last 20% requires debugging. LogPiper gives you structured error data without needing to use a traditional debugger. - **LogPiper: A Universal Logging Bus That Ships Free Inside ToolPiper** (2026-04-05) — https://modelpiper.com/blog/logpiper-universal-logging-for-local-ai ToolPiper includes a real-time logging service any app, script, or MCP tool can write to and query: structured queries, SSE streaming, and correlation IDs. - **Offline Voice Typing on Mac: Dictate Without Internet on Apple Silicon** (2026-04-05) — https://modelpiper.com/blog/offline-voice-typing-mac ToolPiper runs STT on Apple's Neural Engine with zero internet dependency. Push-to-talk and 142 voice commands work offline. Free, no account needed. - **Voice Coding on Mac Without the Cloud: Dictate into Cursor, VS Code, and Terminal** (2026-04-05) — https://modelpiper.com/blog/voice-coding-mac-local Push-to-talk dictation for developers on Apple's Neural Engine. 127-173 ms, private, free. Plus over 420 MCP tools for system control from Claude Code. - **Best Free Dictation App for Mac in 2026: Apple, Whisper, and the On-Device Option** (2026-04-04) — https://modelpiper.com/blog/free-dictation-app-mac Four genuinely free dictation options on Mac. Apple built-in, Whisper.cpp, MacWhisper, and ToolPiper compared for accuracy, latency, and privacy. - **Private Voice Dictation on Mac: No Cloud, No Data Collection, No Compromise** (2026-04-04) — https://modelpiper.com/blog/private-voice-dictation-mac ToolPiper runs Parakeet STT on Apple's Neural Engine. 127-173 ms, 25 languages, fully offline, free with no account. Your voice never leaves your Mac. - **Visual Testing on Mac: Record, Replay, Export to Playwright** (2026-04-04) — https://modelpiper.com/blog/visual-testing-no-code-mac Record browser tests visually, replay with self-healing selectors, export to Playwright or Cypress. No code required. AX-native, local, Chrome-based. - **Wispr Flow Alternative for Mac: Private, On-Device Dictation** (2026-04-04) — https://modelpiper.com/blog/wispr-flow-alternative-free-mac ToolPiper runs STT on Apple's Neural Engine. Free, private, 127-173 ms latency, plus 142 voice commands Wispr doesn't have. - **Best Ollama Frontend for Mac: Every GUI Option Compared** (2026-04-03) — https://modelpiper.com/blog/best-ollama-frontend-mac Every Ollama frontend on Mac compared: Ollama App, Open WebUI, LM Studio, Ollamac Pro, Askimo, plus ToolPiper, which replaces the runner instead of fronting it. - **Browser Health Monitoring: Catch What Tests Miss on Mac** (2026-04-03) — https://modelpiper.com/blog/browser-health-monitoring-tests Passive health monitors run after every test step. Console errors, JS exceptions, HTTP failures - caught automatically with zero configuration. - **Reduce Test Maintenance Cost on Mac: AX Selectors and Self-Healing** (2026-04-03) — https://modelpiper.com/blog/reduce-test-maintenance-cost Hard numbers on E2E test maintenance: 30-50% of testing time goes to upkeep. AX selectors and self-healing change the equation. PiperTest is $49/month. - **Temporal Assertions: Always, Eventually, Next Testing on Mac** (2026-04-03) — https://modelpiper.com/blog/temporal-assertions-testing Three temporal assertion modes that replace brittle waits. Always holds for duration, eventually within deadline, next on immediate check. 100ms polling. - **Test Coverage on Mac: See Which Interactions Your Tests Actually Cover** (2026-04-03) — https://modelpiper.com/blog/test-coverage-visualization-mac Interaction coverage shows which buttons, forms, and links your tests actually touch - not just which code lines execute. PiperProbe scans the AX tree. - **Testim and mabl Alternative: Enterprise Self-Healing Testing for $49/Month** (2026-04-03) — https://modelpiper.com/blog/testim-mabl-alternative-free Enterprise testing tools gate self-healing behind quote-only contracts. PiperTest delivers the same capabilities locally on your Mac for $49/month. - **Accessibility-Native Testing: Why the AX Tree Is the Right Abstraction for Selectors** (2026-04-02) — https://modelpiper.com/blog/accessibility-native-testing-ax-selectors The technical case for testing against Chrome's real accessibility tree instead of the DOM: how AX selectors work, and why getByRole is a DOM simulation. - **Accessibility Testing Automation on Mac: AX Selectors as Audits** (2026-04-02) — https://modelpiper.com/blog/accessibility-testing-automation-ax AX-native test selectors double as accessibility audits. If your app isn't accessible, your tests won't resolve. Free on Mac. - **AI Test Generation on Mac: From AX Snapshot to Running Tests** (2026-04-02) — https://modelpiper.com/blog/ai-test-generation-local Generate browser tests with any AI model. PiperTest feeds the AX tree as plain text, AI reasons about what to test, and tests run locally on Mac. - **Export Tests to Playwright and Cypress From Mac in One Click** (2026-04-02) — https://modelpiper.com/blog/export-tests-playwright-cypress Export PiperTest recordings to clean Playwright or Cypress code. Deterministic selector mapping, mock rendering, temporal comments. Zero vendor lock-in. - **MCP Testing Tools: How AI Agents Write and Run Browser Tests** (2026-04-02) — https://modelpiper.com/blog/mcp-testing-ai-agents 20 MCP tools for AI-driven browser testing. Agents take AX snapshots, generate steps, run them with self-healing, and export to Playwright. Any MCP client. - **Cypress Alternative: Self-Healing AX Selectors That Don't Need the Cloud** (2026-04-01) — https://modelpiper.com/blog/cypress-alternative-self-healing PiperTest vs Cypress: Cloud costs, no multi-tab, Safari still experimental, and cy.prompt() sending page data off-machine. Honest comparison and migration path. - **Fix Flaky Tests on Mac: Self-Healing AX Selectors That Work** (2026-04-01) — https://modelpiper.com/blog/fix-flaky-tests-self-healing Stop chasing brittle CSS selectors. PiperTest targets the accessibility tree and self-heals broken selectors in 5-15ms, locally on your Mac. - **No-Code Browser Testing: The Tool QA Engineers Have Been Waiting For** (2026-04-01) — https://modelpiper.com/blog/no-code-browser-testing-tool You know every user flow in your app, you just cannot write Playwright tests. PiperTest records, replays, and exports browser tests visually. No code needed. - **Playwright Alternative: AX-Native Visual Testing With Self-Healing Selectors** (2026-04-01) — https://modelpiper.com/blog/playwright-alternative-visual-testing PiperTest vs Playwright, compared honestly: real AX tree, self-healing selectors, visual authoring, 20 MCP tools. Browser automation free, PiperTest $49/mo. - **Selenium Alternative 2026: An Honest Look at What's Changed and What Hasn't** (2026-04-01) — https://modelpiper.com/blog/selenium-alternative-2026 Selenium still dominates browser automation, but 45% of teams report constant test breakage. What a modern Selenium alternative actually looks like in 2026. - **Automate Demo Videos on Mac: AI Screenplay to Rendered MP4** (2026-03-31) — https://modelpiper.com/blog/automate-demo-videos-mac Define a screenplay, let AI record and narrate your demo, and render a finished MP4. All local on Mac. No cloud, no editing timeline. - **Mac Screen Capture with Live AI Streaming: Record, GIF, and Stream to Vision Models** (2026-03-31) — https://modelpiper.com/blog/mac-screen-capture-ai-streaming Capture any region of your Mac screen, record to MP4, export as GIF or WebP, and stream live at 30fps to AI vision models - all from a free menu bar app. - **Ollama Chat Without Docker on Mac: Native Alternatives to Open WebUI** (2026-03-31) — https://modelpiper.com/blog/ollama-no-docker-mac Open WebUI needs Docker, eating 2-4GB of RAM before you load a model. ToolPiper is a native Mac app that discovers your Ollama models automatically. - **Playwright vs Cypress vs PiperTest in 2026: The Testing Framework Comparison Nobody Wrote** (2026-03-31) — https://modelpiper.com/blog/playwright-vs-cypress-vs-pipertest Playwright vs Cypress vs PiperTest for 2026: master feature table, MCP tool analysis, real AX tree against DOM injection, performance data, and recommendations. - **Self-Healing Test Selectors: How PiperTest Fixes Broken Tests Automatically** (2026-03-31) — https://modelpiper.com/blog/self-healing-test-selectors How self-healing selectors work: three modes from 5ms AX fuzzy matching to AI-assisted repair, compared with Testim, mabl, Cypress, and Playwright. - **Test Recorder for Browser on Mac: AX-Enriched, Not Brittle** (2026-03-31) — https://modelpiper.com/blog/test-recorder-browser-mac Record browser tests with AX-enriched selectors that self-heal. No browser driver, no brittle CSS. Direct CDP WebSocket, structured steps, one-click export. - **Desktop Automation on Mac: AI Controls Your System Settings, Windows, and Apps** (2026-03-30) — https://modelpiper.com/blog/desktop-automation-ai-mac AI that types and talks but can't move a window or toggle dark mode. ToolPiper closes that gap with 142 native macOS actions across 26 domains, free. - **Hover Zoom Alternative for Chrome: Full-Size Image Preview Without the Privacy Risks** (2026-03-30) — https://modelpiper.com/blog/hover-zoom-alternative-chrome Preview full-size images and videos by hovering over thumbnails, with intelligent CDN pattern discovery and no tracking. For Chrome, Firefox, and Safari. - **Push-to-Talk AI on Mac: Dictate and Command from Any App** (2026-03-30) — https://modelpiper.com/blog/push-to-talk-ai-mac Hold a key, speak, release. Text appears at your cursor or your Mac executes a command. All local, all on Apple Neural Engine, 127-173 ms latency. - **Capture Per-App Audio on Mac Without Virtual Drivers** (2026-03-29) — https://modelpiper.com/blog/capture-app-audio-mac-no-drivers Record and stream audio from any individual Mac app, Chrome, Spotify, or Zoom, without virtual audio drivers. AudioPiper uses Core Audio Taps built into macOS. - **Live Transcription on Mac: Real-Time Speech to Text as You Speak** (2026-03-29) — https://modelpiper.com/blog/live-transcription-mac See words appear on screen as people speak, processed entirely on your Mac. Streaming STT with AudioPiper and FluidAudio on the Neural Engine. - **Mac Clipboard Manager with AI Snippets: One Private App (2026)** (2026-03-29) — https://modelpiper.com/blog/mac-clipboard-manager-ai-snippets A free Mac clipboard manager with AI snippets - 2,000-item history, image OCR, on-device rewrites, and snippet import that auto-detects 8 formats. - **Browser Automation on Mac: Your Local LLM Sees and Controls Chrome** (2026-03-28) — https://modelpiper.com/blog/browser-automation-local-ai-mac AI browser automation that keeps your page content local. ToolPiper's 25 CDP tools let a local LLM drive Chrome via the accessibility tree. - **Code Search with Local RAG on Mac: Ask Your Codebase Without Uploading It** (2026-03-28) — https://modelpiper.com/blog/code-search-rag-local-mac Search your codebase with AI using local RAG on Mac. Semantic code search, hybrid vector + keyword retrieval, all on-device. No code leaves your machine. - **Local AI Agents on Mac: Tool Calling Without the Cloud** (2026-03-28) — https://modelpiper.com/blog/local-ai-agents-mac Run AI agents locally on your Mac with over 420 MCP tools. Tool calling, multi-step reasoning, desktop control - no cloud APIs, no data leaving your machine. - **OLLAMA_ORIGINS=*: Fix the Ollama CORS Error on Mac** (2026-03-28) — https://modelpiper.com/blog/ollama-cors-fix-mac Ollama allows localhost by default, so the CORS error hits other origins. Set OLLAMA_ORIGINS, restart, verify. Or skip it: ToolPiper ships CORS built in. - **Web Scraping on Mac: 7 Extraction Formats with Framework Detection** (2026-03-28) — https://modelpiper.com/blog/web-scraping-local-ai-mac Scrape JavaScript-heavy websites using a real browser, detect 16 frontend frameworks, and extract content in 7 formats. All local, no cloud APIs. - **Apple Intelligence and Open Models on Mac: Use Both in One App** (2026-03-27) — https://modelpiper.com/blog/apple-intelligence-open-models-mac Apple Intelligence runs on the Neural Engine. Open models run on Metal GPU. ToolPiper runs both in one app with smart routing and a shared interface. - **Local Embeddings on Mac: On-Device Vector Search Without Cloud APIs** (2026-03-27) — https://modelpiper.com/blog/local-embeddings-mac Run embedding models and vector search entirely on your Mac. On-device embeddings by default, HNSW indexing, and an OpenAI-compatible API. Zero data exposure. - **Local OpenAI-Compatible API on Mac: Drop-In Replacement on Localhost** (2026-03-27) — https://modelpiper.com/blog/local-openai-api-mac Run an OpenAI-compatible API on localhost with ToolPiper. Same SDK, same code, same prompts. Just change the base URL and your LLM runs locally on Mac. - **Local MCP Server on Mac: over 420 AI Tools in One Install** (2026-03-27) — https://modelpiper.com/blog/mcp-server-local-mac ToolPiper is a 421-tool MCP server for macOS. Local inference, browser automation, voice, vision, desktop control, and testing from one native app. - **AI Model Memory on Mac: How Much RAM Do You Actually Need?** (2026-03-26) — https://modelpiper.com/blog/ai-model-memory-mac How much RAM does a local LLM need on Mac? At Q4, a 7-8B model needs ~5GB, a 14B ~9GB, a 70B ~42GB. Per-model tables for Apple Silicon unified memory. - **Local AI vs ChatGPT Plus: What You Get Without the Subscription** (2026-03-26) — https://modelpiper.com/blog/local-ai-vs-chatgpt-plus ChatGPT Plus costs $240/year. Local AI on your Mac costs nothing per query. Here is what you gain, what you lose, and when each makes sense. - **Local LLM Benchmarks on Apple Silicon: Measured Token Speed on M2 Max** (2026-03-26) — https://modelpiper.com/blog/local-llm-benchmarks-apple-silicon How fast is a local LLM on Apple Silicon? Measured on an M2 Max 32GB: an 8B runs ~28 tok/s at Q4. Bandwidth-scaled estimates for the rest, plus your own chip. - **Which Local LLM on Mac: A Model Selection Guide for Apple Silicon** (2026-03-26) — https://modelpiper.com/blog/which-local-llm-mac Hundreds of models on HuggingFace, one Mac. Here's the decision tree for picking the right local LLM based on your RAM, use case, and quality needs. - **Install MLX-Audio on Mac: Python TTS and the Zero-Code Alternative** (2026-03-25) — https://modelpiper.com/blog/install-mlx-audio-mac How to install and use MLX-Audio for local text-to-speech on Mac - then discover ToolPiper, which bundles the same models in a native app with zero Python required. - **Install Ollama on Mac: Setup Guide and the One-App Alternative** (2026-03-25) — https://modelpiper.com/blog/install-ollama-mac Step-by-step guide to install and configure Ollama on Mac, plus ToolPiper, the native app that runs the same GGUF models with no terminal. Free, no account. - **Pose Estimation and Mocap on Mac: Skeleton Tracking Without the Cloud** (2026-03-20) — https://modelpiper.com/blog/pose-estimation-mocap-mac-local Detect human poses, track skeletons, and stream motion capture in real time on your Mac's Neural Engine. No cloud, no markers, no special hardware. $29/month. - **Local Video Upscaling on Mac: Free 2x Upscale at 44 FPS** (2026-03-19) — https://modelpiper.com/blog/local-video-upscale-mac Free video upscaling on your Mac's Neural Engine: 360p to 720p at 44 FPS, faster than realtime playback. No cloud upload, no watermark, no subscription. - **Local RAG Chat on Mac: Ask Your Documents, Keep Your Data** (2026-03-18) — https://modelpiper.com/blog/local-rag-chat-mac Index your files and ask questions about them using local AI - retrieval-augmented generation running entirely on your Mac, no documents uploaded anywhere. - **Local Image Upscale on Mac: 4x Resolution Without the Cloud** (2026-03-17) — https://modelpiper.com/blog/local-image-upscale-mac Upscale photos and screenshots 2x or 4x on your Mac using CoreML super-resolution models. No upload, no compression loss. Part of ModelPiper Studio ($29/mo). - **Deep Reasoning on Mac: Complex Problem-Solving Without the Cloud** (2026-03-15) — https://modelpiper.com/blog/deep-reasoning-mac-local Reasoning models think step by step before answering. Now they run locally on your Mac - private, free per query, with full chain-of-thought transparency. - **Image Narration on Mac: AI Describes What It Sees and Reads It Aloud** (2026-03-14) — https://modelpiper.com/blog/image-narration-mac-local-ai Drop an image into ModelPiper - a vision model describes what's in it, then text-to-speech reads the description aloud. All on-device, all private. - **Local Document OCR on Mac: Extract Text Without Uploading Anything** (2026-03-13) — https://modelpiper.com/blog/local-document-ocr-mac Extract text from scanned documents, photos, and screenshots using Apple Vision OCR - on-device, fast, and private. No cloud upload required. - **Screen Q&A on Mac: Ask AI About Anything on Your Screen** (2026-03-12) — https://modelpiper.com/blog/screen-qa-visionpiper Select any region of your screen, ask a question, get an answer - all locally. VisionPiper captures your screen and feeds it to a vision model running on your Mac. - **Voice Cloning on Mac: Replicate Any Voice, Entirely Local** (2026-03-11) — https://modelpiper.com/blog/voice-cloning-mac-local Clone any voice from a short audio sample, entirely on your Mac - no biometric data uploaded anywhere. Voice is too sensitive for the cloud. - **Live Translation on Mac: Speak One Language, Hear Another** (2026-03-10) — https://modelpiper.com/blog/live-translation-mac-local Real-time speech translation running entirely on your Mac - speak English, hear Portuguese. No cloud, no Google Translate, no data leaving your machine. - **Transcribe and Summarize on Mac: Audio In, Key Points Out** (2026-03-08) — https://modelpiper.com/blog/transcribe-summarize-mac Drop a meeting recording into ModelPiper and get a structured summary of decisions, action items, and key points. No confidential audio uploaded anywhere. - **Voice Chat on Mac: Talk to AI Locally, Hear It Respond** (2026-03-07) — https://modelpiper.com/blog/voice-chat-mac-local-ai A full voice conversation with AI, speech-to-text through language model to text-to-speech, running entirely on your Mac. No cloud, no latency, no data leaving. - **macOS TTS Voices: Local Text to Speech on Mac Without the Cloud** (2026-03-06) — https://modelpiper.com/blog/local-text-to-speech-mac Three neural TTS voices that run on your Mac's own GPU. No cloud service ever sees your text, no per-character billing, and no character limit. - **Transcribe Audio on Mac Free: Local Voice Transcription, No Cloud** (2026-03-05) — https://modelpiper.com/blog/local-voice-transcription-mac Transcribe audio files on your Mac for free with Whisper-class accuracy. Meetings, lectures, and voice memos, entirely offline, with no audio uploaded anywhere. - **Private Local Chat on Mac: ChatGPT Without the Cloud** (2026-03-04) — https://modelpiper.com/blog/private-local-chat-mac Run a private AI chatbot entirely on your Mac. Prompts and chats stay on your machine, no API keys, no internet. How local LLM chat works on Apple Silicon. - **Local-First AI on macOS: Why Your Data Should Never Leave Your Machine** (2026-03-03) — https://modelpiper.com/blog/local-first-ai-macos Your Mac has dedicated AI hardware built in. Why local-first AI matters: privacy by architecture, zero API costs, no rate limits. How ModelPiper delivers it. ## Papers > 9 entries below. Each row is a one-line summary; fetch the linked `.md` for the full body. - **Eight WebMCP Tools That Registered Nothing: What Mocked Tests Hide** (2026-08-01) — https://modelpiper.com/blog/webmcp-chrome-api-migration-mocked-tests We shipped eight WebMCP tools that registered nothing for three months. Every test stayed green, because they stubbed the browser API. The bug class, and the fix. - **Why PiperKit Exists: Local AI Is All That's Left** (2026-04-25) — https://modelpiper.com/blog/why-piperkit-exists PiperKit LLC builds local AI for Apple Silicon. Cloud AI lost the trust argument, open source closed the quality gap, the next chips finish the job. - **Your AI Coding Agent Needs a Log Store, Not Terminal Output** (2026-04-07) — https://modelpiper.com/blog/ai-agent-feedback-loop-log-store The write-run-check-fix loop breaks when check means reading terminal scrollback. A structured log store turns AI debugging from guessing into querying. - **An MCP Tool That IS a Logging Service: How LogPiper Closes the Agent Feedback Loop** (2026-04-07) — https://modelpiper.com/blog/mcp-logging-tool-logpiper-architecture Every MCP observability tool monitors agents from outside. LogPiper is a logging service the agent uses directly - bidirectional, persistent, and queryable. - **Vibe Debugging: The Observability Gap in AI-Assisted Development** (2026-04-05) — https://modelpiper.com/blog/vibe-debugging-observability-ai-development AI coding assistants write code but can't see runtime behavior. LogPiper gives them a queryable log store - the missing feedback loop for agentic development. - **AX-Native Browser Automation: Why We Built Our Own CDP Engine** (2026-03-24) — https://modelpiper.com/blog/ax-native-browser-automation-cdp-engine How ToolPiper's accessibility-tree-first browser automation enables AI-powered test generation that works with any model provider, not just MCP-aware clients. - **PiperSR: First Open-Source Super-Resolution Model for Apple Neural Engine** (2026-03-24) — https://modelpiper.com/blog/pipersr-open-source-ane-super-resolution PiperSR is a 453K-parameter super-resolution model built for Apple Neural Engine. 37.54 dB PSNR on Set5, 44 FPS video upscale. CoreML, open source, any Mac. - **Building Over 420 MCP Tools in Swift: Architecture of a Unified Local AI Server** (2026-03-22) — https://modelpiper.com/blog/mcp-tools-swift How ToolPiper became the first MCP server to unify LLM inference, TTS, STT, embeddings, OCR, vision, browser automation, and RAG behind one install, in Swift. - **How We Achieved 44 FPS Video Upscale on Apple Neural Engine** (2026-03-21) — https://modelpiper.com/blog/pipersr-44fps-video-upscale-apple-neural-engine A deep dive into PiperSR's double-buffered ANE+Metal pipeline that upscales 360p video to 720p at 44.4 FPS - 1.5x realtime on Apple Silicon.