Skip to main content
The AI Copilot is a chat assistant for investigation. Ask it about a spike, an incident, or a service, and it queries your streams, traces, and incidents on your behalf — then explains what it found. It runs against the LLM provider you configure, and every conversation is auditable.

​
Configure a provider

Under Settings → AI providers, add a provider and pick a model:
ProviderNotes
AnthropicClaude models.
OpenAIGPT models.
OpenAI-compatibleAny endpoint that speaks the OpenAI API (self-hosted, gateways).
API keys are sealed at rest with the instance envelope key. The Copilot is off until a provider is configured.

​
Prompts

Settings → AI prompts holds the system prompts the Copilot uses. Edit them to match your environment — naming conventions, which streams matter, your incident workflow — so answers stay grounded in how your team operates.

​
Toolsets

The Copilot answers by calling tools, not by guessing. A toolset declares which read-only capabilities the model may use — for example, running a query, listing streams, or fetching incidents — exposed over an MCP-style dispatcher. Configure them under Settings → AI toolsets to scope what the assistant can reach.
Tools are read-only investigation helpers. The Copilot inspects your data to explain an anomaly; it does not change configuration or write streams.

​
Audit

Every Copilot session, message, and tool call is recorded. Review them under Settings → Audit — who asked what, which tools ran, and what came back — so AI-assisted investigations are as accountable as any other action.

Model selection

Pick the latest, most capable models from your provider for the best investigation quality.