devin-2api

English | 中文

devin-2api is a lightweight forwarding tool for the OpenAI Responses API. It exposes a standard /v1/responses endpoint and transparently forwards your LLM requests to Devin (app.devin.ai) through an adapter — letting external programs call Devin's models through the standard OpenAI protocol.

Features

• OpenAI-compatible /v1/responses endpoint — call Devin's models through the standard OpenAI Responses protocol
• Streaming and non-streaming responses (typed SSE / JSON)
• Adapter-based design — easily extended to new upstreams
• Easy to deploy — single static binary, public Docker image on Docker Hub
• Optional debug logs per request for troubleshooting

Quick start

1. Get a Devin token

devin-2api authenticates to Devin with your Devin session token. On macOS, extract it from the Devin app's local state:

sqlite3 ~/Library/"Application Support"/Devin/User/globalStorage/state.vscdb \
  "SELECT json_extract(value, '$.apiKey') FROM ItemTable WHERE key='windsurfAuthStatus';"

The output is a token in the devin-session-token$... format.

2. Configure

cp config.example.yaml config.yaml

Edit config.yaml and fill in your token (starting from config.example.yaml, you only need to fill in devin.token — the base URL and model are pre-filled as examples).

3. Run

Local:

go run ./cmd/devin-2api -config config.yaml

Docker (image published on Docker Hub):

docker run --rm -p 8080:8080 \
  -v "$PWD/config.yaml:/app/config.yaml" \
  leokun123/devin-2api --config /app/config.yaml

4. Verify

curl http://localhost:8080/healthz
# {"status":"ok"}

Usage

Note: /v1/* endpoints support optional API key authentication. Set auth.api_key in config.yaml to require clients to send Authorization: Bearer <api_key> or X-Api-Key: <api_key>. If left empty, the endpoints remain open (only expose them to trusted networks).

Call http://localhost:8080/v1/responses with your OpenAI Responses API client.

Non-streaming:

curl http://localhost:8080/v1/responses \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5-2",
    "input": "Hello"
  }'

Streaming (SSE):

curl -N http://localhost:8080/v1/responses \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5-2",
    "input": "Hello",
    "stream": true
  }'

The request body follows the OpenAI Responses API (input, instructions, tools, stream, …). See the Contributing guide for the exact subset of fields supported.

Configuration

Configuration is a YAML file loaded once at startup. Unknown fields are rejected.

Field	Description	Required
server.listen	HTTP listen address	Yes
devin.base_url	Devin Connect service base URL	Yes, once devin.token is set (no default in code; config.example.yaml uses https://server.codeium.com)
devin.token	Devin session token (devin-session-token$...)	No — endpoint returns 503 until set
devin.model	Devin chat model UID (e.g. glm-5-2)	Yes, once devin.token is set (no default in code)
debug.enabled	Write per-request debug logs under logs/ next to the config file	No
auth.api_key	API key for /v1/* endpoints; empty disables auth. Clients may send Authorization: Bearer <key> or X-Api-Key: <key>	No

server:
  listen: ":8080"

devin:
  base_url: "https://server.codeium.com"
  token: "devin-session-token$..."
  model: "glm-5-2"

debug:
  enabled: false

auth:
  # Set to a strong key to protect /v1/*; leave empty to keep endpoints open.
  api_key: ""

Notes:

• tokens are never written to logs (redacted as <redacted>);
• if devin.token is empty, /v1/responses returns 503 provider_configuration;
• config.yaml is tracked by git — don't commit a real token (add it to .gitignore if needed).

Documentation

• Architecture, supported API fields, proto extraction, and other technical details: Contributing guide
• License: MIT