Belay – See what keeps going wrong in your Claude Code and Codex sessions

Doplex Labs launched Belay, a local tool that scans existing Claude Code and Codex session histories to flag loops, stalls, and failed commands, then proposes reviewed rules and verification commands for the next run. In a preprint evaluation of approved lessons delivered to agents, tasks were completed in 11 of 16 cases with Belay versus 7 of 16 with a human-written instruction and 6 of 16 with no additional guidance. Belay stores session history, evidence, and guidance in an encrypted local database, requires no account, and does not upload data to Doplex Labs.

Across the sessions, the agent said done while the typecheck still failed. Belay groups the repeated pattern and keeps the proof attached. For the next run, your agent receives a proposed rule and verification command. You review it before it has any authority. Built for the agents you already use. Claude Code Codex Install once. Start with the sessions you already have. 1 Your history, in one place. Belay finds the coding sessions saved on your machine and loads them into a local timeline. You don't need to start a new run to find something useful. Loading your session history Claude Code Codex ~/.belay/timeline.html Your sessions. On your machine. ✓ Existing history loaded. New sessions added as you work. 2 Find what kept going wrong. Belay flags loops, stalls, and failed commands across your history. Open a session to see the work that went nowhere, and where you had to step in. Billing webhooksSession timeline 14:02 Edited six files across the APIpackages/api/src/billing/ 14:31 Ran the unit tests. All passed.pnpm test billing → 48 passed 14:33 Marked the task complete. Never ran the typecheck.pnpm typecheck → 14 errors, left behind 15:10 You stepped in."the build is red again" Third session this week to end this way. Now your agent can see that. 3 See what it cost you. And what to say instead. Your own agent debriefs each session: where your prompting or scoping cost time and money, and the exact message to send next time. Billing webhooks · DebriefWritten by your Claude Code You named the files on your fourth message. Before that, the agent searched the whole repo three times. 13:38–14:02 searching for the files14:02 edits and tests Scopingabout 24 min · roughly $6 What you did Asked to “fix the webhook retry bug” with no file names. The agent grepped three times and read 40 files before you pasted the paths. What an expert would have done Named the two files and the failing test in the first message. You knew where the bug lived. Say this insteadThe retry bug is in packages/api/src/billing/webhooks.ts and retry.ts. The failing test is webhooks.test.ts. Fix it there, no need to search the rest of the repo. 1 of 2 things to change · 1 to keepStays on this machine 4 Have your agent fix the cause. Ask your agent to look at the failed sessions. It reads Belay's evidence through MCP and proposes a real fix: a hook, a rule, a project instruction. You review it before anything changes. Your coding agentConnected to Belay Why do my sessions keep ending with a broken build? Three times this week I marked work done without running the typecheck. Nothing in this project stops me. I can add a hook that runs it before I'm allowed to finish. Belay prepares reviewed guidance with the evidence, scope, and verification command attached. Your next agent starts with the lesson that matters instead of rediscovering it from scratch. Mission Pack · typecheck before done 1 relevant lesson ready for the next session Original evidence attachedReady Scope and limits includedReady Verification command includedReady Preprint results Lessons from past work helped on new tasks. Tasks completed when a prior lesson was useful, across Claude Code and Codex. 11/ 16Belay 7/ 16Human-written instruction 6/ 16No additional guidance This evaluation measured approved lessons delivered to agents; session debriefs have not been evaluated yet. Broader evaluation is underway; full methodology and agent-level results will follow in a research preprint. Belay stores session history, evidence, and guidance in an encrypted local database and does not upload it to Doplex Labs. No account required. Read the trust details. ↳ Local collection. Local storage. Belay reads the session files already on your machine and keeps its records there. It works offline. ↳ Your agent stays your agent. Belay has no model. Optional analysis uses your configured Claude Code or Codex provider, under the account and policies you already use. ↳ Your work keeps moving. Belay never blocks an agent action. If Belay stops, your agent keeps going. Belay for Teams · Enterprise Coming soon. One Belay across every machine at your company. Fleet-wide patterns One fix for everyone Local-first, like solo Join the waitlist Company Make the next run a little wiser. Less going in circles. More getting somewhere. For you, and for your agent.