Skip to main content

​
Logs and memory

Termyte keeps its stable check history and explicit user memory inside the current repository. 
.termyte/logs.jsonl
.termyte/memory.jsonl

​
Review check decisions

termyte logs
termyte logs --blocked
termyte logs --warned
termyte logs --agent codex
termyte logs --today
termyte logs --json
Each check event includes the redacted command, decision, risk, reason, matched rules, policy sources, memory matches, and optional agent/session context. termyte check writes these events. policy test, logs, and memory do not write check events.

​
Mark exact commands safe or unsafe

termyte mark-unsafe "npm test"
termyte mark-safe "npm run generated-code-cleanup"
termyte memory
Memory is:
  • repository scoped;
  • explicitly created by you;
  • matched against normalized exact command text;
  • stored after command redaction.
An unsafe exact match can upgrade allow to warn. Safe memory does not weaken a warning, ask, or block.

​
Runtime ledger and semantic memory

The experimental run -- and shell surfaces use a separate SQLite ledger and semantic memory in .termyte/termyte.db. 
termyte replay
termyte inspect -- npm publish
The runtime ledger records planned and finalized outcomes. Runtime semantic memory aggregates repeated action outcomes and adds context to explanations; it does not currently weaken runtime policy.

​
Treat local state as sensitive

Termyte redacts recognized secret values and stores environment variable keys instead of values. Redaction is pattern based, so review local operational state before sharing it.