LibreChat custom endpoint for an OpenAI-compatible gateway: YAML, .env and silent drops

[apigoto]

I work on the APIGOTO team. APIGOTO is an independent OpenAI-compatible gateway, not affiliated with OpenAI, Anthropic, Google or LibreChat.

0. Check the endpoint first

curl https://api.china-aiapi.com/v1/models -H "Authorization: Bearer $KEY"

1. .env

APIGOTO_API_KEY=your-key

2. librechat.yaml (keep your existing version: line)

endpoints:
  custom:
    - name: 'APIGOTO'
      apiKey: '${APIGOTO_API_KEY}'
      baseURL: 'https://api.china-aiapi.com/v1'
      models:
        default: ['MODEL_ID']
        fetch: true
      titleConvo: true
      titleModel: 'current_model'
      modelDisplayLabel: 'APIGOTO'

To let each user bring their own key, use apiKey: 'user_provided'. LibreChat asks for it on first use and stores it encrypted.

3. Docker: mount the file

Copy docker-compose.override.yml.example to docker-compose.override.yml and uncomment the bind mount of ./librechat.yaml to /app/librechat.yaml. Then:

docker compose down && docker compose up -d

It should appear in the endpoint selector.

Troubleshooting: the endpoint never appears

LibreChat keeps a custom endpoint only if name, baseURL, apiKey and models are all present, and models has fetch: true or a non-empty default. A block that fails this check is dropped with no error and no log line. Check:

• Spelling of those four keys, and that the block sits inside endpoints.custom.
• name is unique. A second endpoint with the same name (case-insensitive) silently replaces the first.
• Schema errors anywhere in librechat.yaml stop the server entirely. Run docker compose logs api.

Other errors

• Missing API Key for APIGOTO: the ${APIGOTO_API_KEY} variable isn't in .env, or the container wasn't restarted.
• 404s: baseURL should end at /v1, not /v1/chat/completions.
• Model list empty with fetch: true: some endpoints don't serve /models to every key. Put the IDs in default and set fetch: false.
• A model rejects a parameter: list it under dropParams (the docs example drops stop).

We use one endpoint for several model families, with upstream failover (at most 3 retries) and a per-request log. Pay-as-you-go, BYOK supported: www.china-aiapi.com

For people running several custom endpoints: do you set titleModel to a cheap fixed model, or leave it on current_model?