Skip to content
Agentmine

Redaction & privacy

Agentmine redacts high-confidence secret patterns before storing searchable message text and keeps all data local by default. Lossless source records remain sensitive.

What gets redacted

Normalization redacts high-confidence secret patterns before storing searchable text. Built-in patterns cover:

  • common API keys
  • bearer tokens
  • private keys
  • OAuth-style tokens
  • Slack tokens
  • AWS access key IDs
  • GitHub token prefixes
  • secret-shaped environment values

Disabling redaction

Use --no-redact only for a deliberate local audit where preserving exact text is required. This flag controls canonical session fields; workflow records described below remain sensitive in either mode.

Lossless records remain unredacted

Redaction protects normalized searchable session text and canonical message and tool previews. It does not make every bounded value in the database safe. Lossless JSON and full-output fields may retain secrets: this includes raw_events.raw_json, tool_outputs.output_text, tool_calls.args_json, and message_parts.payload_json.

Workflow ingest is a separate lossless path and does not pass through canonical session redaction. raw_workflow_runs.raw_json, raw_workflow_runs.raw_path, and raw_workflow_journal.raw_json retain source values. Derived fields copied from those records — including workflow_runs.summary, workflow_runs.script_path, workflow_run_phases.detail, workflow_agents.result_preview, and workflow_agents.result_full — may also retain secrets. The agentmine workflow command bounds the result_full excerpt size but does not redact it.

Synced raw transcript archives also keep their original content. Protect the data directory, sessions.db, command output, and backup archives as sensitive local data; do not publish or share them without a separate review.

Local-first data stance

Agentmine is local-first. It reads local transcript stores and writes local SQLite data under the user data directory by default (see Data paths). It does not call an LLM in the default sync -> normalize -> extract path, so session content does not leave the machine during ordinary ingest.