Skip to content

oai2ollama is a lightweight FastAPI server that translates an OpenAI ChatCompletions API endpoint into an Ollama-compatible endpoint. It allows you to use custom LLM providers for coding agents that don't support custom OpenAI APIs but do support Ollama (like GitHub Copilot for VS Code).

Quick Start

with Python

You can run directly via uvx (if you have uv installed) or pipx:

uvx oai2ollama --help

usage: oai2ollama [--api-key str] [--base-url HttpUrl] [--capabilities list[str]] [--models list[str]] [--host str]
options:
  --help, -h                    Show this help message and exit
  --api-key str                 API key for authentication (required)
  --base-url HttpUrl            Base URL for the OpenAI-compatible API (required)
  --capabilities, -c list[str]  Extra capabilities to mark the model as supporting
  --models, -m list[str]        Extra models to include in the /api/tags response
  --host str                    IP / hostname for the API server (default: localhost)

Or you can use a .env file:

OPENAI_API_KEY=your_api_key
OPENAI_BASE_URL=your_base_url
HOST=0.0.0.0
CAPABILITIES=["vision","thinking"]
MODELS=["custom-model1","custom-model2"]

To mark the model as supporting certain capabilities, use the --capabilities (or -c) option:

oai2ollama -c tools or oai2ollama --capabilities tools

oai2ollama -c tools -c vision or oai2ollama --capabilities -c tools,vision

To support models that are not returned by the /models endpoint:

oai2ollama -m model1 -m model2 or oai2ollama -m model1,model2

Capabilities currently used by Ollama are: tools, insert, vision, embedding, thinking and completion. We always include completion.

with Docker

docker build -t oai2ollama .
docker run -p 11434:11434 \
  -e OPENAI_API_KEY="your_api_key" \
  -e OPENAI_BASE_URL="your_base_url" \
  oai2ollama

To listen on all interfaces:

docker run -p 11434:11434 oai2ollama --host "::"

Features

  • OpenAI-compatible /v1/chat/completions endpoint
  • Ollama-compatible /api/tags and /api/show endpoints
  • Streaming response support
  • Configurable model list with extra models
  • Customizable capabilities (tools, insert, vision, embedding, thinking)