AIgent Squad Chat — Configuration¶

There are two places to configure, with a clean separation of concerns:

Provider (secret) → the Grafana LLM app. URL + token, never in this plugin.
Behaviour knobs → this plugin's Configuration page. No secrets.

1. Provider — in the Grafana LLM app¶

Configure AIgent Squad as a custom OpenAI-compatible provider in the LLM app settings (secureJsonData), or via provisioning:

apiVersion: 1
apps:
  - type: grafana-llm-app
    org_id: 1
    disabled: false
    jsonData:
      openAI:
        provider: openai
        url: https://aigent-squad.<domain>   # NO /v1 suffix — the app appends it
    secureJsonData:
      openAIKey: $__env{AIGENT_SQUAD_TOKEN}

  - type: staffops-aigent-squad-chat-app
    org_id: 1
    disabled: false

2. Plugin behaviour — Configuration page¶

Administration → Plugins → AIgent Squad Chat → Configuration. Persisted in the plugin's jsonData (no secrets):

Setting	Default	Purpose
Model	`base`	Model slot requested from the LLM app (the squad routes by context).
Source tag	`[source:grafana-plugin/aigent-squad-chat]`	Prefix on the latest user message for audit/filtering. Empty disables it.
Thinking indicator delay (ms)	`3000`	Idle time before the "working…" indicator.
Max stream idle (ms)	`120000`	Hard timeout if no chunk arrives.
Expand tool trace by default	`off`	Whether the reasoning panel starts expanded.

Defaults live in src/config.ts, so the plugin works with no configuration.

Local end-to-end stack¶

The package ships a docker-compose.e2e.yaml (Grafana 13.x + grafana-llm-app + the built plugin) plus provisioning. Provide the token via env:

export AIGENT_SQUAD_TOKEN=<token>     # e.g. from your secret store
docker compose -f docker-compose.e2e.yaml up -d
# open http://localhost:3000 → AIgent Squad Chat