AI Observability API - Squasher

Use the AI Observability API for aggregate dashboard views over LLM requests, costs, latency, sessions, users, and hosted fix activity. Use Agent Observations API for row-level generation, tool-call, retrieval, and agent-step evidence.

Operations

Operation	Method and path	Purpose
`ai.stats`	`GET /v1/projects/{project_id}/ai/stats`	Aggregate AI request and session stats.
`ai.histogram`	`GET /v1/projects/{project_id}/ai/histogram`	Request, cost, latency, and token buckets over time.
`ai.getSession`	`GET /v1/projects/{project_id}/ai/sessions/{session_id}`	Session drill-down.
`ai.getUser`	`GET /v1/projects/{project_id}/ai/users/{user_id}`	User drill-down, optionally bounded by `from` and `to`.
`ai.fixStats`	`GET /v1/projects/{project_id}/ai/fixes/stats`	Hosted fix request and completion breakdowns.
`ai.fixTimeline`	`GET /v1/projects/{project_id}/ai/fixes/timeline`	Hosted fix dispatch and completion counts over time.

Keep dashboard queries bounded by time when supported. Use observation filters when you need the exact prompt, output, tool call, token count, or cost row.

CLI and MCP

Use squasher ai ..., squasher agent-observations ..., or squasher fixes ... depending on the workflow. In MCP Code Mode, call search("ai") or search("agentObservations") first.

Agent handoff

Analyze AI usage for Squasher project <project_id>. Start with ai.stats or ai.histogram, then drill into ai.getSession, ai.getUser, or agentObservations only when a concrete session, user, model, tool, or cost spike is identified. Return ids, timestamps, model/provider names, tokens, costs, and latency.

Related: AI observability, Agent SDK.