Data policy

What FeedGPT stores, who processes it, and for how long. Last updated 30 October 2026.

What FeedGPT stores

The records your connectors send to a source's webhook, as sent, and the fields FeedGPT extracts from them. Your sources, mappings, templates and settings. The names, email addresses and roles of the people in your organization. A log of API calls (which endpoint, when, and the response status, not the body).

Who can read it

People in your organization, and the API keys and connected AI clients you create. A key can be narrowed to some sources, and every read is limited to what it may see. FeedGPT staff read your data only to fix a problem you report, or when the law requires it.

What leaves FeedGPT

FeedGPT uses these services to run. Each receives only what its job needs.

  • Vercel

    Hosts the FeedGPT application and its API. Records pass through it on the way to the database.

  • Neon

    The Postgres database (AWS, us-east-1) where your sources, records and settings are stored.

  • Trigger.dev

    Runs background work: reprocessing, fixes, deliveries and agent runs. It receives the id of the job, never your records.

  • Clerk

    Sign-in and organization membership: names, email addresses and roles of the people in your organization.

  • Modal

    Runs the language models FeedGPT deploys: schema inspection, and agents when your organization turns them on. Records an inspection or an agent reads are sent to that model to be answered.

  • A model API for mapping suggestions

    Scores suggested fixes for a field. It receives field names and types, and record values only if your organization chooses to send them.

  • Upstash

    Rate-limit counters. It holds counts keyed by organization, never records.

  • Sentry

    Error reports. Request bodies, payloads and record values are removed before a report is sent.

  • Polar

    Billing and subscriptions: your billing email and plan. Never records.

Your records also go wherever you send them: the AI clients you connect over MCP or the API, and the endpoints you add as destinations.

Agents

Agents are off until an organization admin turns on Agents may read records. While it is on, the records an agent reads from the sources you give it are sent to a model FeedGPT runs on Modal, to write the agent's report. Nothing is sent until an agent runs. Turning it off pauses every agent and stops every run at its next step; records a run already read are not recalled.

Training

FeedGPT does not use your records to train models, and does not sell or share them for advertising.

How long data is kept

Records and their raw payloadsUntil you delete them, or delete the source.
A deleted sourceRestorable for 7 days, then removed with its records.
Records sent to your own systemsThe delivery log is kept 15 days.
Alert deliveries to your own systemsAttempts 30 days, deliveries 90 days.
Agent runs and their reportsFor the life of your organization. Revoking an agent stops it and keeps its history.
Audit history (mapping changes, setting changes)For the life of your organization.

Deleting your data

Delete a record or a source from the dashboard. To delete your whole organization and everything in it, contact us.

Questions

Ask through the contact page. If this policy changes, the date at the top changes with it.