opencode-plugboard

Appearance

Langflow ​

Langflow is a visual builder for agents and flows. Its language model components use providers you set in the UI or with env vars.

Upstream: source, docs.

Plugboard options ​

Give Langflow its own key. Docker needs a reachable address (Serving other machines).

jsonc
{
  "options": {
    "listen": { "host": "0.0.0.0", "port": 4010 },
    "keys": { "langflow": "{env:PLUGBOARD_KEY_LANGFLOW}" },
    "models": ["opencode-go/glm-5.3-flash"],
    "aliases": { "chat": "opencode-go/glm-5.3-flash" },
    "apps": { "langflow": { "allowedModels": ["chat"] } }
  }
}

Keep the app's allowedModels list short. Langflow lists everything /v1/models returns for the key.

Connect ​

Set these in Langflow's environment:

sh
OPENAI_COMPATIBLE_BASE_URL=http://host.docker.internal:4010/v1
OPENAI_COMPATIBLE_API_KEY=<the langflow key>
LANGFLOW_SSRF_ALLOWED_HOSTS=host.docker.internal

Langflow blocks private addresses by default, so the allowlist entry is required. Add it before you save the base URL.

Or use the UI: profile menu, Settings, Model Providers, OpenAI Compatible, then set Base URL and API Key. Langflow calls /v1/models to validate the endpoint and discover models. Enable the models you want, then pick OpenAI Compatible and the model in a Language Model component.

What works ​

  • Language Model components, through the chat alias.
  • Agents and flows that use tools. Plugboard passes tools through.

Limits ​

  • Embedding models appear under Embedding Models after discovery, but plugboard serves none. Use an embedding provider you run.
  • If the base URL is accepted but no models appear, check the models and apps policy.