Insights & Observability
The Observe section gives you eyes on what your agents and deployments are actually doing — catching regressions before and after they ship, and understanding how well your agents follow your team’s conventions. It groups three surfaces: Regression Radar, Shadow Twin, and Team Genome.
Regression Radar
What it is: A post-deployment monitor that spots regressions by lining up service health against your deploy events. Access via Observe → Regression Radar in the sidebar.
How it works: Pick a time range (24h, 7d, 30d) and Regression Radar plots your services’ health over time with deploy markers overlaid — so when something degrades, you can see which deploy lines up with it.
- Service health overview — every service with its status (green / yellow / red), error rate, latency, and trend.
- Deploy timeline — deploy events placed on a timeline; click one for details and the metrics it affected.
- Deploy–metric correlator — links degraded metrics to the deploy that likely caused them.
- Auto-rollback history — automatic rollbacks with the reason, recovery status, and time to recover.
Shadow Twin
What it is: A pre-deployment safety check that runs a candidate build alongside production and compares how the two behave — before you ship. Access via Observe → Shadow Twin in the sidebar.
How it works: Shadow Twin exercises the new build against the same workload as production and diffs the results, then gives you a clear go / no-go recommendation.
- Summary — endpoints tested, regressions found, improvements, and what stayed the same.
- Verdict — a color-coded recommendation (Safe to deploy, Review recommended, or Do not deploy) with a confidence score.
- Behavior diff — endpoint-by-endpoint comparison of production versus the candidate, with the delta and a status badge. Filter to show only regressions or only improvements.
- Regression detail — deeper context on anything that got worse.
Team Genome
What it is: A profile of your team’s coding conventions and a score for how closely each agent follows them. Access via Observe → Team Genome in the sidebar.
How it works: Team Genome scans your codebase for recurring patterns — naming, structure, architecture, testing — builds a profile of your conventions, and scores each agent’s adherence. You can rebuild the profile any time to rescan.
- Team alignment — a radar chart of how well the team aligns across convention dimensions.
- Convention confidence — overall confidence based on how much code was analyzed, with a rebuild button.
- Detected conventions — the conventions found, grouped by category, with confidence and how often each appears.
- Agent alignment — per-agent alignment scores and a drift score that flags agents straying from the norm.
- Drift alerts — callouts for agents with significant drift and which convention they’re deviating from.
See also: Cockpit, Cortex, Runs & Orchestration