Anything AI.
On your Mac.

Everything Ollama does, free. Then it does the rest of your Mac.

ModelPiper is a suite of local-first AI apps for macOS. ToolPiper, the core app, downloads models, runs them on the native llama.cpp engine, and serves a local OpenAI-compatible API — no account, no caps, no cloud. Then add voice, vision, automation, and over 420 MCP tools on top.

Download ToolPiper Open Web App

Free · Apple Silicon · macOS 26+

Always Free

The Pipeline Builder

A visual canvas for building AI workflows. Drag, connect, and run — right in your browser.

ModelPiper pipeline builder showing a 4-block workflow with Text, RAG, and AI Provider blocks alongside the My Connections panel listing local models loaded in ToolPiper.

Block-Based

Snap together prompt, model, response, and vision blocks to create multi-step AI pipelines.

Any Model

Works with Ollama, OpenAI, Anthropic, Groq, and any OpenAI-compatible endpoint.

Browser-Based

No installation needed. Open the web app and start building immediately.

Free Beta

ToolPiper for macOS

The local AI engine for your Mac. One click to download models, run them on the embedded llama.cpp engine, and read unified logs — free, no account.

ToolPiper Menu Bar Screenshot

Curated Model Library

Browse and install models with one click. LLMs, vision models, and TTS engines — all pre-configured.

Bundled Inference

llama.cpp, Apple Intelligence, speech-to-text, and text-to-speech — all running on your hardware.

Unified Logs

Every HTTP request, response, and error from all backends in one searchable stream.

Zero Config

Download a model, launch, and go. ToolPiper handles inference, model management, and server lifecycle.

Download for macOS

Requires macOS 26 or later

Coming Soon

VisionPiper

Real-time camera and screen analysis powered by local vision models.

  • Live camera feed analysis with local vision models
  • Screen capture and OCR pipelines
  • Custom vision workflows with the pipeline builder
  • Runs entirely on-device — no cloud uploads
VisionPiper Screenshot

From the Blog

Technical writing on local AI, MCP servers, on-device inference, and AX-native browser testing.

Accessibility-Native Testing: Why the AX Tree Is the Right Abstraction for Selectors

The technical case for testing against Chrome's real accessibility tree instead of the DOM.

AX-Native Browser Automation: Why We Built Our Own CDP Engine

A custom Chrome DevTools Protocol engine that queries the AX tree directly. Self-healing in 5 to 15 ms.

Give Claude over 420 macOS Superpowers with One MCP Server

ToolPiper exposes over 420 local tools to any MCP-compatible AI agent. One local server, no cloud.

Apple Intelligence vs Open Models on Mac

When the Foundation Models Framework wins and when llama.cpp wins. Routing strategy that uses both.

The AI Agent Feedback Loop Is Broken. Here's the Fix.

Why terminal output is write-only memory for AI coding agents, and what a queryable log store changes.

Local AI Agents on Mac: Build and Run Without the Cloud

A tour of the local-first agent stack: inference, tools, memory, and orchestration on Apple Silicon.

Run an MCP Server Locally on Mac

Setup guide for the loopback HTTP transport, with Claude Code, Cursor, and Windsurf integration.

Run a Local OpenAI-Compatible API on Mac

A drop-in replacement for the OpenAI API that runs entirely on your machine.

View all 152 posts →

How It Works

Three steps to your first AI pipeline.

1

Install ToolPiper

Download the macOS menu bar app. Pick a model and it downloads — ready to run on your hardware.

2

Open the Web App

Launch the pipeline builder in your browser. Add blocks, pick a model, and wire them together.

3

Run Your Pipeline

Hit run. Your prompts flow through connected models — streaming responses in real time.

Your Data Stays on Your Mac

ModelPiper runs on your machine. Your prompts, files, responses, and workflows stay on your device — unless you pick a cloud provider. The only thing it ever sends on its own is an optional anonymous benchmark score you choose to run, published to a public leaderboard.

Explore plans

Free for everyone. Pro for the full toolkit. Studio for creators, Max for developers. Team for your whole office on one Mac.

Free

Everything Ollama does, free. No account, no caps.

$0

  • Native llama.cpp engine — run any GGUF model
  • Unlimited model downloads, multi-model switching
  • Local OpenAI-compatible API + embeddings
  • MCP server with 359 free tools
  • All speech: transcription, text-to-speech, voice cloning, dictation
  • Chat with bundled model
  • Apple Intelligence on the Neural Engine
  • Full browser automation, vision, and system control
  • Visual pipeline builder
  • Free companion apps (VisionPiper, AudioPiper, MediaPiper)
Most popular

Pro

What no model runner does, at any price.

$10/mo

  • Everything in Free
  • Local RAG over your files
  • Web scraping and YouTube transcripts
  • Cloud API proxy (bring your own keys)

Studio

For content creators and marketers.

$29/mo

  • Everything in Pro
  • Image upscaling (ANE-native)
  • Video upscaling (60fps real-time)
  • Video editing pipeline
  • Pose detection (60fps streaming)
  • Outreach toolkit (queue, posting, Firehose)

Max

For developers and QA engineers.

$49/mo

  • Everything in Studio
  • Code (agentic AI code editor)
  • PiperTest (self-healing browser tests)
  • API discovery toolkit
  • Priority support
For teams

Team

One Mac runs your AI. Your whole team uses it — governed, audited, private.

$99/mo per deployment

  • Everything in Max, on the deployment
  • Unlimited named member tokens — no per-seat pricing
  • Per-user attributed audit trail with pull export
  • Remote tool governance from modelpiper.com
  • Clients on any OS via piper-bridge
  • Priced per deployment — add Macs as you grow

Common Questions

What ModelPiper runs, what it costs, and what never leaves your Mac.

Does ModelPiper work offline?

Yes. Chat, inference, speech, vision, and macOS automation all run on-device through the bundled llama.cpp Metal engine and Apple Intelligence. The only time ModelPiper reaches the network is when you download a model or opt into publishing an anonymous benchmark score. No account is required to run it.

Which models are bundled with ModelPiper?

The engines ship with the app, not the weights. ToolPiper bundles the llama.cpp Metal runtime, an in-process MLX-Swift peer, speech-to-text, and text-to-speech. Models install in one click from a curated library of GGUF LLMs, vision models, and TTS voices, and any GGUF model you supply also works.

How much does ModelPiper Pro cost?

Pro is $10/mo. The Free tier has no account and no caps: the native llama.cpp engine, unlimited model downloads, a local OpenAI-compatible API, the MCP server with 359 free tools, and all speech features. Pro adds local RAG over your files, web scraping, YouTube transcripts, and a bring-your-own-key cloud proxy.

Which macOS versions does ModelPiper support?

ToolPiper requires macOS 26 or later on Apple Silicon — an M1 chip or newer. Inference runs on the Metal GPU and the Apple Neural Engine, so Apple Silicon is a requirement rather than a recommendation. The visual pipeline builder at modelpiper.com runs in any modern browser, on any operating system.

Does any of my data leave my Mac?

No. Your prompts, files, responses, and workflows stay on your machine. Three things reach the network, each only when you start it: browsing or downloading a model queries HuggingFace, which can log your IP and search terms; a cloud provider you pick with your own API key; and an anonymous benchmark score you publish.

How do I connect ModelPiper to Claude Code or another MCP client?

Run: claude mcp add --transport http toolpiper http://127.0.0.1:9998/mcp. ToolPiper exposes a local MCP server over streamable HTTP on loopback, and the server ships inside the app, so there is no separate binary to install. Cursor, Windsurf, and any other streamable-HTTP MCP client connect to the same URL.

Can I use ModelPiper with my existing OpenAI or Ollama tooling?

Yes. ModelPiper serves a local OpenAI-compatible API, so anything already pointed at OpenAI or Ollama works after you change the base URL. You can also bring your own Anthropic, OpenAI, or Groq keys and route them through the Pro cloud proxy, which stores every key in the macOS Keychain.

Get Notified When v1 Launches

Be the first to know when ToolPiper leaves beta and VisionPiper drops.