About Retrack
Retrack builds experience for AI agents: memory earned from verified runs, recalled only when it applies, protected from poisoning and measured.
Updated
Retrack builds experience for AI agents.
Production agents run the same kinds of tasks thousands of times, yet each run starts from zero. A person who did a job a thousand times would have learned its shortcuts, its traps and its exceptions. An agent has not. Retrack gives agents that experience: it captures every run, checks which ones worked, turns them into procedures and warnings, and hands the right ones back before the next similar task.
What we believe
- Memory must be earned. Only verified runs teach. A run that merely looked successful must never become a procedure.
- Wrong memory is worse than none. Memory is recalled only when it clearly applies.
- Memory is an attack surface. Anything an outsider could have influenced is checked or reviewed before it can change what an agent does.
- Claims must be measured. Every project has a holdout group, and every number we publish comes from a run that can be shown, including the numbers that did not improve.
Who builds it
Retrack is built by Gokulakrishnan Jawahar, an independent developer in Chennai, India. It is in early access, and feedback from teams running agents in production shapes what comes next.
How it is built
- The connector,
retrack-ai, is open source under the MIT licence and published on PyPI. - The service runs on Vercel in Mumbai with a Postgres database and vector search in the same region; Claude reads runs to judge outcomes, write playbooks and check for injections.
- Each project's data is isolated by row-level security, and secrets are redacted before anything is stored.
Contact
Write to retrackit@gmail.com. Read the privacy policy and terms.