Privacy Policy
Last updated 3 November 2026
Retrack (retrack-ai.online) gives AI agents experience memory. This policy explains what data Retrack handles, why, who helps process it, and the choices you have. Retrack is in early access.
Who runs Retrack
Retrack is operated by Gokulakrishnan Jawahar, Chennai, Tamil Nadu, India. For anything about your data, including requests and complaints, write to retrackit@gmail.com. That address is also the grievance contact under India's Digital Personal Data Protection Act, 2023.
What we collect
Your account: when you sign in with GitHub, GitHub shares your name, email address, username and profile picture with us. We never get access to your code or repositories.
Your agents' runs, sent by the retrack-ai connector you install: the task text, each tool the agent called with its inputs and a shortened result (up to 1,000 characters each), the final answer (up to 4,000 characters), token counts, cost, latency, the model and tool versions, and any outcome you report with rt.feedback. Obvious secrets such as API keys, tokens and passwords are removed on your machine before anything is sent, and removed again on our servers.
What Retrack learns from those runs: playbooks, failure warnings, regression test cases, and numeric representations of task text (embeddings) used to find similar tasks.
Service records: API keys (we store only a one-way hash and a short prefix), recall logs, and daily usage counters.
How we use it
Only to run Retrack for you: to judge whether runs succeeded, turn them into playbooks and warnings, check them with the memory firewall, recall them for your agent, and show them in your dashboard. We also use usage records to keep the service working and within its limits.
Your data is never sold, never used for advertising, and not used to train AI models, by us or by the AI providers we use.
Who helps us process it
Supabase stores the database and handles sign-in (Amazon Web Services, Mumbai, India).
Vercel hosts the website and API (Mumbai, India).
Anthropic's Claude reads run records to judge outcomes, write playbooks and check for prompt injection (United States). Its API does not train on this data.
Google Cloud (Vertex AI) turns task text into embeddings (Google's global endpoint, which may process data outside India). Paid Vertex AI does not train on this data.
GitHub provides sign-in.
Each of them handles data only to provide its service to us. Some processing therefore happens outside India.
How long we keep it
Runs are deleted automatically after your project's retention period: 90 days unless you change it in your project's settings. Playbooks, warnings and test cases stay until you retire them or ask us to delete them. Your account and projects stay until you ask us to delete them.
Security
All traffic is encrypted (HTTPS). Every project's data is isolated in the database, so one project can never read another's. API keys are stored only as hashes. The memory firewall keeps untrusted or suspicious runs out of active memory until they are checked or you approve them.
Cookies
We use only the cookies needed to keep you signed in. There are no advertising or analytics trackers.
Your choices and rights
You can see your runs and memory in the dashboard, retire playbooks, revoke API keys and change retention at any time. To get a copy of your data, correct it, or delete your account and all your projects, write to retrackit@gmail.com; we reply within 30 days. You may also withdraw consent by stopping the connector and asking us to delete your data. If you are in India, you may complain to the Data Protection Board of India; if you are in the EU or UK, to your data protection authority.
Please do not send health, financial account or other sensitive personal data through your agent's runs unless you need to and are allowed to.
Children
Retrack is for developers and businesses, not for anyone under 18.
Changes
If this policy changes, we update this page and its date, and email you about significant changes before they apply.
Questions? Write to retrackit@gmail.com.