AIgent Squad Chat¶
- Plugin ID:
staffops-aigent-squad-chat-app - Type: app
- Package:
packages/staffops-aigent-squad-chat-app
A Grafana app plugin that provides a conversational UI to AIgent Squad — an external SRE-agentic system (OpenAI-compatible API, specialist subagents). The plugin is a thin chat client: it streams the conversation and renders it; AIgent Squad owns all orchestration and tool execution (Grafana / Loki / Tempo / PromQL / Kubernetes) server-side.
graph LR
U[User] --> UI["Chat page (React)"]
UI -->|"grafana/llm"| LLM["Grafana LLM app"]
LLM -->|"OpenAI chat/completions"| SQUAD["AIgent Squad"]
SQUAD --> TOOLS[("Loki / Tempo / PromQL / K8s")]Features¶
| Feature | Description |
|---|---|
| Streaming chat | Tokens render as they arrive over Grafana Live. |
| Thinking indicator | "AIgent Squad is working…" while a tool runs server-side. |
| Multi-turn | Full conversation history is replayed each turn. |
| Reasoning panel | The <think> tool-trace is parsed into a collapsed panel, separate from the answer. See Reasoning & Trace. |
| Stop / retry | Cancel an in-flight stream or re-run after an error, preserving partial content. |
| Availability gate | Degrades gracefully when the Grafana LLM app is missing/unconfigured. |
| Configuration page | Model, source tag, timeouts, reasoning default — no secrets. See Configuration. |
Requirements¶
- Grafana 13.x (declared
grafanaDependency >= 12.3.0). - The Grafana LLM app (
grafana-llm-app) installed and enabled, with a provider configured to the AIgent Squad endpoint. It is a declared plugin dependency. - Access restricted to Editor role and above.
Architecture (Option A)¶
The provider URL + token live in the Grafana LLM app (secureJsonData); the browser never
sees them. See Architecture for the full rationale (Option A vs a
self-contained Go backend) and the guardrail-compatibility decision.