2ollama

2ollama

Proxy server that translates various LLM API formats to Ollama.

Use case: Run local LLM completions with tools that only support cloud APIs (like Zed editor's Codestral integration).

Supported Providers

Provider	Endpoints	Status
Codestral	/v1/fim/completions, /v1/models	✅
OpenAI	/v1/chat/completions	🔜 Planned
Anthropic	/v1/messages	🔜 Planned

Installation

npm install -g 2ollama

Or run directly with npx:

npx 2ollama

Usage

Start the proxy

2ollama

Options

-p, --port <PORT>          Port to listen on (default: 8787)
-o, --ollama-url <URL>     Ollama server URL (default: http://localhost:11434)
-m, --model <MODEL>        Default model to use (default: codestral:latest)
-d, --daemon               Run in background
-h, --help                 Show help
-v, --version              Show version

Environment variables

PORT=8787
OLLAMA_URL=http://localhost:11434
DEFAULT_MODEL=codestral:latest

Examples

# Use a different model
2ollama --model qwen2.5-coder:7b

# Run on a different port
2ollama --port 8080

# Run in background
2ollama --daemon

# Connect to remote Ollama
2ollama --ollama-url http://192.168.1.100:11434

Zed Configuration (Codestral)

• Start the proxy: 2ollama

• Configure Zed's settings.json:

{
  "language_models": {
    "codestral": {
      "api_url": "http://localhost:8787"
    }
  },
  "features": {
    "edit_prediction_provider": "codestral"
  }
}

• Enter any string as the API key when prompted (it's ignored by the proxy)

Recommended Models

Any Ollama model with FIM (fill-in-middle) support works. For instance:

• codestral:latest - Mistral's code model (22B, best quality)
• qwen2.5-coder:7b - Good balance of speed and quality
• qwen2.5-coder:1.5b - Fast, lower resource usage
• deepseek-coder-v2:16b - Strong coding performance
• starcoder2:7b - Solid alternative

Pull your preferred model:

ollama pull qwen2.5-coder:7b

API Endpoints

Endpoint	Method	Description
/	GET	Health check
/health	GET	Health check
/v1/fim/completions	POST	FIM completions (Codestral format)
/v1/models	GET	List available models

Programmatic Usage

import { startServer } from "2ollama";

startServer({
  port: 8787,
  ollamaUrl: "http://localhost:11434",
  defaultModel: "codestral:latest",
});

Architecture

2ollama uses a pluggable provider system. Each provider defines routes that translate incoming API requests to Ollama's format.

src/
  provider.ts           # Provider interface
  providers/
    index.ts            # Provider registry
    codestral/          # Codestral provider
      index.ts          # Route handlers
      types.ts          # Request/Response types
      transform.ts      # Format transformations

Adding a Provider

import type { Provider } from "2ollama";

export const myProvider: Provider = {
  name: "my-provider",
  routes: [
    {
      method: "POST",
      path: "/v1/my/endpoint",
      handler: async (req, res, ctx) => {
        // Transform request, call Ollama, return response
      },
    },
  ],
};

Development

# Install dependencies
bun install

# Run in development mode
bun run dev

# Build
bun run build

# Type check
bun run typecheck

# Lint/format
bun run check

License

MIT