HIVE

What Hive captures — and what it doesn't

Hive is a shared team memory that makes our AI tools collectively smarter. Its primary consumer is the tools, not people-watching.

Captured automatically

• Git commit / PR descriptions from connected repositories
• Claude Code session summaries (the closing recap, not the full transcript)
• Concise summaries of exchanges in Claude Desktop and GitHub Copilot — what was built, decided, or answered (the outcome, not the full conversation)
• Search queries run against Hive (topic + time, to know what's current)

Where your data goes

The Hive service, its database, and search run on our own infrastructure. Embeddings are the exception: to make text searchable, the captured text above is sent to a hosted AI provider through our LiteLLM gateway to be turned into search vectors. It is used only for that, not to train any provider's models — but it does leave our network for that step. A deployment can avoid this by using a local embedding model instead.

Never done

• No team-wide productivity or trend dashboard
• No attribution of failures to individuals
• No person-related feature without your explicit opt-in

Your controls

• Opt in / out of person-related features (e.g. expert routing): POST /api/consent/<you> {"opted_in": true|false}
• Export your data (GDPR): GET /api/gdpr/<you>/export
• Delete your data (GDPR): DELETE /api/gdpr/<you> — your knowledge entries stay (they're team knowledge) but are un-attributed; your query log, trigger events, and consent record are removed.

Expert routing stays fully disabled until works-council sign-off (see SIGN-OFF.md).