Dashboard
What each view shows and how the shared filters scope them.
After garcon setup, the empty dashboard guides you to Settings → Connect a harness. Choose your tool and account label, copy its configuration, restart the tool and send one short request. The dashboard refreshes automatically when usage arrives. Verify the first request.
http://127.0.0.1:4141. Usage views share time-range, harness, account and (once sync has pulled in a second machine) device filters; Settings always describes the whole instance.
- Overview, Usage, Models, Accounts
- Request and token volume, composition, rankings.
- Cost
- Estimated spend at published pay-as-you-go list prices (never a bill), cache savings, blended $/1M tokens, run rate, an editable pricing table whose edits stay in the browser.
- Sessions
- Requests grouped into coding sessions per account, harness and device, with a timeline and per-session tokens, peak context and cost.
- Activity
- Hour-of-day and calendar heatmaps, streaks, peak concurrency.
- Performance, Latency
- Generation speed and time-to-first-byte by model; the proxy's own connection overhead versus upstream time.
- Logs
- Sortable, filterable request table with CSV export; a Device column appears while sync is on.
- Settings
- How the instance is wired, a snippet builder for connecting harnesses, the sync switch and its status, browser-side state. For another machine, reuse the project URL and secret key with a distinct device name; see the join instructions.
If the dashboard becomes unreachable, run garcon doctor. After an update, garcon doctor --wait 10s can wait for the new version to start. Sync is optional; leaving it off still lets this machine record and display usage.