Skip to main content
Voss records harness sessions so work can be resumed and audited. This page covers single-agent harness sessions. A team run additionally persists a session tree — one node per agent — plus per-card review sidecars, which form the run’s audit trail.

​
List sessions

The session list shows saved work with enough context to choose what to resume.

​
Resume a session

Resume restores prior conversation context and model/session metadata. Use voss sessions first when you need the id. In the Go terminal client, /resume opens a picker that shows and searches each session’s first prompt, and /resume <id-or-name> or voss ui resume <id-or-name> resumes directly. Saved messages reappear in the transcript. Use /save-session [name] to name a conversation so it is easy to find later.

​
Task journal for server clients

When a client drives Voss through voss serve, task state and events are also written to a SQLite journal at .voss/task-journal.sqlite3. Clients can replay every event after a cursor, so a dropped connection or a client restart does not lose history. If the server stops mid-task, the work is marked interrupted on the next start; it is never relaunched automatically.

​
What sessions are for

  • Continuing multi-step repo work
  • Preserving task goals and assumptions
  • Auditing tool use and decisions
  • Avoiding repeated context reconstruction

​
Redaction rule

Provider credentials and API keys must not be written into session payloads.
Do not paste secrets into natural-language prompts. Voss avoids storing provider credentials, but user-provided text can still become part of a transcript.