Mining
Usual reads selected local Claude and Codex conversation files when you explicitly ask it to learn from them.
Usual keeps its own history, retrieval, and review on your computer. There is no Usual account, hosted inference service, or subscription to cancel.
That does not mean your coding model is offline. It means Usual does not add another cloud service to the middle of your workflow. Your existing Claude Code or Codex session still handles the reasoning under its normal provider policies.
Usual reads selected local Claude and Codex conversation files when you explicitly ask it to learn from them.
The default decision database is ~/.usual/judgment.sqlite3, outside your project repositories.
Relevant evidence is selected locally for the current coding session. Original transcripts are not modified.
The decision review UI binds to your own computer. Accepted or corrected choices stay part of your local workflow.
When the current agent reads selected excerpts, that agent's provider processes them according to its normal data handling. Usual makes no model API calls of its own and does not claim to make Claude or Codex inference offline.
Local does not mean invisible. Review the evidence your agent is about to use, and follow the privacy controls and policies of the coding client you already use.
The public site serves the landing page, documentation, source bundle, and worked examples. It does not expose a private judgment corpus or a public transcript-ingestion endpoint.
Common secrets are redacted before persistence, but Usual's redaction is not complete PII detection. The safe habit is still to review what you select and what your agent sees.
See the full privacy and data-boundary notes, or start with the installation guide.