Letta Proxy

OpenAI-compatible proxy for Letta Cloud agents.

Point any OpenAI-compatible client (OpenClaw, LangChain, OpenAI SDK, etc.) at this proxy and it will route requests to your Letta agents. Includes a web dashboard for testing.

Features

• Drop-in OpenAI API — /v1/models and /v1/chat/completions (streaming supported)
• Full tool-calling — bidirectional tool call round-trips
• Web dashboard — manage API keys, list agents, test chat in-browser
• Two modes:
  • dashboard — full web UI + API (default)
  • light — API only, minimal dependencies
• Docker support — docker run ready

Quick Start

# 1. Clone
git clone https://github.com/kaivyy/lettaproxy.git
cd lettaproxy

# 2. Start (auto-installs dependencies)
./start.sh

# 3. Open dashboard and enter your Letta API key
#    http://localhost:1416

That's it. The dashboard will guide you through connecting your Letta account.

Alternative: Set key via environment variable

export LETTA_API_KEY="***"
./start.sh

Light mode (API only, no dashboard)

./start.sh light

Configuration

Variable	Default	Description
LETTA_API_KEY	(empty)	Letta Cloud API key. Get one at app.letta.com
LETTA_BASE_URL	https://api.letta.com	Letta API base URL (change for self-hosted)
PROXY_PORT	1416	Proxy listen port
PROXY_HOST	0.0.0.0	Proxy bind address

API Endpoints

OpenAI-Compatible

Method	Path	Description
GET	/v1/models	List Letta agents as OpenAI models
POST	/v1/chat/completions	Chat with a Letta agent (streaming supported)

Dashboard (dashboard mode only)

Method	Path	Description
GET	/	Web dashboard UI
GET	/api/status	Connection & agent status
POST	/api/set-key	Save Letta API key (validates before saving)
POST	/api/chat	Simple test chat

Both modes

Method	Path	Description
GET	/health	Health check

Tool Calling

The proxy supports full bidirectional tool calling:

1. Client sends tools[] in OpenAI function-calling format
2. Proxy forwards tools to the Letta agent
3. Agent may call tools → proxy returns tool_calls with finish_reason: "tool_calls"
4. Client executes tools and sends results back as tool role messages
5. Agent produces final text response with finish_reason: "stop"

Using with OpenClaw

Add to your openclaw.json under models.providers:

{
  "letta": {
    "baseUrl": "http://127.0.0.1:1416/v1",
    "api": "openai-completions",
    "models": [
      {
        "id": "<agent-id>",
        "name": "My Letta Agent",
        "reasoning": false,
        "input": ["text"],
        "contextWindow": 128000,
        "maxTokens": 8192
      }
    ]
  }
}

Then add the model to your agent's allowlist and use /model letta/<agent-id> to switch.

Docker

docker build -t letta-proxy .
docker run -p 1416:1416 -e LETTA_API_KEY="***" letta-proxy

Self-Hosted Letta

If you're running Letta locally (not Letta Cloud), set the base URL:

export LETTA_BASE_URL=http://localhost:8283
./start.sh

Requirements

• Python 3.9+
• See requirements.txt for dependencies (auto-installed by start.sh)

License

MIT