Runs

A run is a single end-to-end agent execution. Every prompt you send, every skill you invoke, and every multi-agent orchestration produces one or more runs.

Run anatomy

Run
├── Step 1: model call        (prompt → completion)
├── Step 2: tool call         (tool_name, args → result)
├── Step 3: model call        (updated context → completion)
├── Step 4: tool call
└── Step 5: model call        (final answer)

Each step is recorded with:

• Timestamp and duration
• Full input/output payload
• Token counts (for model steps)
• Error details (if the step failed)

Run storage

Runs are stored in SQLite at ~/Library/Application Support/Invoked/invoked.db. The schema is stable and you can query it directly:

SELECT id, summary, model, created_at, duration_ms
FROM runs
ORDER BY created_at DESC
LIMIT 20;

Run metadata

Field	Description
id	UUID, stable across sessions
summary	First 120 chars of the prompt
model	Model used for the run
steps	Ordered array of run steps
status	running | complete | error | cancelled
duration_ms	Total wall-clock time
created_at	ISO 8601 timestamp

Filtering and search

In the Runs panel, use the search bar to filter by:

• Free text (matches summary)
• Model (model:claude-3-5-sonnet)
• Status (status:error)
• Date range (after:2024-01-01)

Programmatic access

Runs are accessible via the IPC API if you're building tooling on top of Invoked.