Insights & telemetry.
The post-sprint debrief: what your agents found, what your runs cost, where quality is trending — and the loop that turns all of it into the next improvement.
01The Insights view
Insights in the left rail has three sections:
02Findings → Compound
Triage is built to end in action, not a list. Work through the findings, select the ones worth acting on, and launch a Compound run seeded with exactly that selection — it fixes what it can directly, promotes what deserves to be a task, and resolves each finding as fixed, triaged, or promoted. One loop from “an agent noticed something” to “the system got better.”
03Code review eval
Sprint and Ship runs can score their own output: at the human-review step, a three-judge Opus jury independently grades the run’s diff against a code-quality rubric, and the score summary renders alongside the review so you’re not approving blind.
- On by default — toggle globally in Settings → Code Review Eval, or per run in the launch wizard’s Advanced options.
- Weak dimensions land as findings, feeding the same triage loop above.
04Telemetry & privacy
Packaged builds ship with anonymized, opt-out telemetry: error reporting (Sentry) and usage metrics (Aptabase). Both toggles live in Settings → Privacy & Telemetry and take effect after a restart.
- Never sent: source code, prompts, file paths, repo names. Every error payload is scrubbed before it leaves your machine.
- No identity: the only identifier is a random install id minted once on your machine.
- Usage metrics report only from release builds; development builds stay silent by default.