RetrySight Docs get-started
What is RetrySight?
Fleet observability for AI coding agents — retries, token waste, and session timelines across Lite, Cloud, and Enterprise.
On this page
RetrySight turns local coding-agent logs into retry timelines, cost reports, and fleet health. Collectors watch tools such as Cursor, Claude Code, and Codex. A manager stores the events. A dashboard (or the Lite desktop app) is how you inspect them.
There is no SDK to drop into your application code. Collectors tail logs that the coding agents already write on each developer machine.
Three ways to run it
| Product | Best for | Where data lives |
|---|---|---|
| RetrySight Lite | Individual developers | On the local machine, loopback only |
| RetrySight Cloud | Small and medium teams | Hosted at cloud-staging.retrysight.com (staging) |
| RetrySight Enterprise | Platform and security teams | Your network — Docker or Helm |
Lite, Cloud, and Enterprise share the same retry taxonomy. They differ in where the manager runs and which collectors are enabled (Enterprise tiers; Cloud uses the full Cloud profile).
What you see
- Retries — edit loops, test failures, rejected diffs, compaction, and subagent spins
- Tokens and cost — usage by model, user, and tool (Cloud Cost page; Enterprise Analytics+)
- Fleet health — collector heartbeats so dashboards do not go quiet
- Session replay — prompt-ready exports, plus MCP on Lite, Cloud, and Enterprise Integration+
Next step
If you are choosing a product, start with Lite vs Cloud vs Enterprise. If you already know, jump to Lite quickstart, Cloud quickstart, or Enterprise Core quickstart.