---
title: "Running ToolPiper alongside Ollama"
description: "Can I run ToolPiper and Ollama at the same time, and will they conflict?"
date: 2026-09-01
author: "ModelPiper"
tags: ["Ollama", "Backends", "Ports", "Compatibility"]
type: "faq"
canonical: "https://modelpiper.com/help/ollama-coexistence"
---

# Running ToolPiper alongside Ollama

> Can I run ToolPiper and Ollama at the same time, and will they conflict?

Yes. ToolPiper and Ollama run side by side. ToolPiper serves on its own port, keeps its own model directory, and never writes into Ollama's. What it does do is read Ollama's library, so a model you have already pulled is not downloaded twice, and, when Ollama is not running, answer Ollama's own API on Ollama's port, so tools built for Ollama keep working.

## Ports

ToolPiper serves on port 9998. Ollama serves on port 11434. By default ToolPiper also answers Ollama's dialect on 11434, but only when that port is free: if Ollama is already running, ToolPiper leaves the port alone and nothing changes for Ollama or for the tools pointed at it. Ollama starts at login on most machines, so it usually gets there first. The listener has its own card in ToolPiper's Settings under General, which shows whether it took the port and turns it off if you would rather it never tried.

## Using models you already pulled

ToolPiper reads Ollama's library on disk, whether or not Ollama is running, and offers to adopt what it finds, with the count and the size on the button. Adoption links each file into ToolPiper's own store without copying it, so a library of many gigabytes is adopted in an instant and takes no extra space. Nothing is written into Ollama's directory, and Ollama keeps working exactly as before.

Removing an adopted model in ToolPiper removes ToolPiper's entry only; Ollama's copy stays. Removing it in Ollama leaves ToolPiper's entry working too.

## Memory is the real constraint

The thing that will actually bite is RAM, not ports. Two large models loaded at once on the same machine compete for memory, and the second one to load may fall back to a slower path or fail to load at all.

If inference gets slow after you start using both, unload the model you are not using. Unloading is immediate and does not delete anything from disk.

## Which one to use for what

Both run GGUF models through llama.cpp on Metal. What ToolPiper adds is its own engines: MLX, and Core ML on the Neural Engine, which Ollama does not use. Use whichever has the model you want loaded; a model adopted from Ollama runs in ToolPiper exactly like one it downloaded itself.