Mattermost AI plugin

The Mattermost AI plugin adds AI assistants (agents) to Mattermost. Each agent uses a service, and the OpenAI Compatible service takes any endpoint that speaks the OpenAI API.

Upstream: source.

Plugboard options

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

{
  "options": {
    "listen": { "host": "0.0.0.0", "port": 4010 },
    "keys": { "mattermost": "{env:PLUGBOARD_KEY_MATTERMOST}" },
    "models": ["opencode-go/glm-5.3-flash", "anthropic/claude-sonnet-4-6"],
    "aliases": { "chat": "anthropic/claude-sonnet-4-6" },
    "apps": { "mattermost": { "allowedModels": ["chat"] } }
  }
}

Connect

In the plugin's service settings, choose OpenAI Compatible in the AI Service dropdown, then set:

Field	Value
API URL	http://host.docker.internal:4010/v1, or the address your Mattermost server reaches (with /v1)
API Key	the mattermost key
Organization ID	empty
Use Responses API	on (the default). It uses /v1/responses
Model	chat

What works

Chat and agents through the Responses API.

Limits

• If agent turns fail, turn Use Responses API off. The agent then uses Chat Completions on the same base URL.
• The plugin's semantic search uses a pgvector store. If it needs an embedding model, plugboard does not serve one, so use another provider for it.