Observability integrations

Observability integrations connect Coldtea to your signal providers: the services that hold your errors, analytics, session replay, observability, and deployment data.

These connections feed self-driving. The signal scout reads connected providers and proposes work backed by what it finds there, so the errors, analytics, and deploys you already track turn into caught regressions instead of one more dashboard to watch.

Providers you can connect

Seven providers connect today, each read-only:

• Sentry for error tracking.
• PostHog for product analytics.
• Honeycomb for observability.
• Datadog for infrastructure, logs, and APM.
• Vercel for hosting and deployments.
• LangWatch for LLM tracing.
• FullStory for session replay.

Sentry connects through OAuth. Honeycomb connects with a configuration key. The others connect from their own settings card. Each connection is read-only: Coldtea reads signals, it does not write to these providers.

Providers Coldtea recognizes but does not connect yet

When the codebase mapper reads a repository, it can recognize dozens more providers than the seven above, across error tracking, analytics, feature flags, alerting, uptime, and logging. Those are planned, not working. Coldtea can tell you it saw one of them in the code, but it cannot read from it yet. A recognized-but-unconnected provider shows up as a request rather than a live connection.

How these feed self-driving

A connected provider becomes a source the signal scout can read on the next daily run. Its findings arrive as tasks with the source signals attached, so you can trace a filed task back to the error, metric, or deployment that prompted it. Connect a provider before a run if you want that run to see it.

For the self-driving side of this, including the connection screen and how signals reach the scout, see connect signals.