Security and privacy.
Crossfire handles documents that are often confidential. This page says what we store, what we send to model providers, what we never do, and how to delete your data.
What we store
- The account link your identity provider issues when you sign in
- The documents you upload to a session
- The findings, rounds, context and decisions in that session
- Usage and cost records per session
We do not store provider tokens.
What we send to model providers
To run a round, the session's documents and objective are sent to Anthropic, OpenAI and Google. The moderator step sends the three analyses labeled A, B and C in place of provider names.
None of the three providers trains on it. Anthropic and OpenAI exclude API content from training by default. Google does not use paid-tier prompts and responses to improve its products. How long each provider keeps API inputs and outputs, and for what purposes, depends on that provider's terms and on the account tier and settings in use. The terms linked above govern.
What we never do
- Sell your documents or share them with third parties beyond the model calls a review requires
- Train models on your documents
- Edit your document without you (Rewrite returns a new, rewritten copy; your original is untouched)
Access
You sign in through our identity provider on the web and in Word. Sessions are private to the account that created them.
Deleting your data
Delete any session from your session list. Deletion is permanent and removes its documents, findings, rounds and history.
Infrastructure
Crossfire runs on Railway behind Cloudflare, with a PostgreSQL database and a durable event log. Traffic is encrypted in transit.