Onyx

Onyx (formerly Danswer) is an enterprise search and chat app. Connectors index your documents, and chat answers with an LLM over that index.

Upstream: source, docs.

Plugboard options

Plugboard serves the LLM role. Indexing and search need an embedding model, which plugboard does not serve, so Onyx uses its own local model or a cloud provider for that.

Role	Served by
LLM (chat and answers)	Plugboard, through the OpenAI-Compatible provider
Embedding (indexing and search)	Onyx's own model server (nomic-ai/nomic-embed-text-v1 by default), or a cloud provider
Reranker	Optional, not served by plugboard

Give Onyx its own key and a text alias. Docker needs a reachable address (Serving other machines).

{
  "options": {
    "listen": { "host": "0.0.0.0", "port": 4010 },
    "keys": { "onyx": "{env:PLUGBOARD_KEY_ONYX}" },
    "models": ["opencode-go/glm-5.3-flash"],
    "aliases": { "text": "opencode-go/glm-5.3-flash" },
    "apps": { "onyx": { "allowedModels": ["text"] } }
  }
}

An app key's model list shows only that app's allowedModels, so text appears in Onyx's list.

Connect

1. Open Language Models in the admin area (/admin/language-models) and choose OpenAI-Compatible.
2. Set the API base to http://host.docker.internal:4010/v1. Onyx adds /v1 if it is missing and reads the model list from <base>/models.
3. Enter the onyx key in the API key field. Leave it set: with an empty field Onyx sends a placeholder, and plugboard refuses it.
4. Fetch the models, select text, and mark it as the default LLM.

What works

Chat and answers through the text alias.

Limits

• Plugboard cannot be the embedding provider. Keep Onyx's default local model, add a cloud provider under Index Settings (/admin/index-settings), or point Onyx at a LiteLLM proxy or another gateway with /v1/embeddings.
• Onyx runs two model servers, each with a Hugging Face cache. Plan disk and network for the first start.
• Plugboard takes data: image URLs by default. It fetches http(s) image URLs only with remoteImages on.
• If text is missing from the model list, check the app's allowedModels list.