docsreliability

Resume the rig. Not the chaos.

Local checkpoints bring your whole multi-terminal grid back — slots, roles, commands, and agent conversations. A Resume Plan shows exactly what will run before anything starts.

overview

Rig Resume restores an entire multi-terminal rig from the latest local Rig Checkpoint. 13x periodically snapshots the active workspace — slots, roles, startup and agent-resume commands, view mode, focus, context-badge state and git metadata — to JSON on disk, then rebuilds and relaunches each slot according to its Resume Mode. It is checkpoint-based: it does not hibernate OS processes, it restarts each slot deterministically and can re-open agent conversations. Before anything runs, a Resume Plan shows exactly what each slot will do, with dangerous commands gated behind explicit confirmation.

  • Local-only: snapshots, latest.json, audit log and scrollback all live under .13x/ in your project
  • No network calls for resume — no code, commands or snapshots ever uploaded
  • Resume Plan previews every command; dangerous ones require explicit confirmation
  • Backend re-validates before stopping any PTY — a rejected resume never touches running terminals
  • Reopens the exact Claude conversation via claude --resume <session-id>, no picker
  • Secrets redacted twice; env vars and .env contents never stored in a checkpoint
why it matters

Plan before run

Every resume produces a Resume Plan first — slots, command previews, git comparison and warnings — shown in a dialog before a single process starts. The backend re-validates selections and dangerous confirmations BEFORE stopping any running terminals, so a rejected resume never touches a live PTY.

Agent-aware

Slots are enriched from agent-launch markers and inferred agent kind (claude/codex/cursor/gemini). Claude chats opened through 13x are relaunched with claude --resume <session-id> to reload the exact conversation with no picker; Codex gets its own resume strategy.

Secrets redacted twice

Commands and scrollback are redacted in the frontend snapshot build and again server-side before any file is written — API_KEY/SECRET/TOKEN/PASSWORD assignments, Bearer tokens, and sk-/ghp- keys. 13x never stores env vars or .env contents in a checkpoint.

Auditable evidence

Every snapshot save, plan build, rejected resume, executed resume and deletion appends a JSON Lines entry to .13x/snapshots/audit.jsonl — selected slots, confirmed dangerous slots, per-slot failures, warnings and the 13x app version.

Visual-only restore

Resume any slot as read-only history: persisted scrollback is painted into a synthetic 'RESTORED' card with no process started — review a previous session without rerunning anything.

Local-first, no telemetry

Snapshots, the latest checkpoint, scrollback and the audit log all live under .13x/ in your project. 13x makes no network calls for resume and uploads no code, commands or snapshots — everything is read and written on the local filesystem.

how it works
01

Capture automatically

13x snapshots the active rig on lifecycle events — rig launch, slot start/stop, kill-all, manual save, agent session changes, context-packet sends — plus a debounced 30s timer and an app-close flush. Secrets are redacted before the snapshot is built.

02

Persist to disk

The backend validates the snapshot, re-redacts secrets, stamps the app version, captures git branch/commit/dirty state off the main thread, and writes <id>.json plus latest.json atomically under .13x/snapshots/ — then prunes old checkpoints by a retention policy.

03

Review the Resume Plan

Click RESUME. When nothing needs a decision, resume runs silently. A running grid, a dangerous command, a Rerun-Last-Command slot, or a changed git branch routes to the Resume Dialog with a per-slot plan, command previews and warnings.

04

Confirm and execute

Select slots, confirm any dangerous or rerun rows, and consent to replacing a running grid. The backend re-validates everything; only then does it stop the grid and launch selected slots staggered by 150ms, replaying agent and restore lines.

05

Rehydrate the grid

Fresh sessions are adopted into the workspace: context badges restored, scrollback painted into non-agent cards, read-only cards synthesized for visual-only slots, view and focus restored — and a fresh checkpoint saved.

faq

Does Rig Resume hibernate my processes?

No. It is checkpoint-based, not process hibernation. 13x restarts each slot from its configured Resume Mode (Resume Agent, Rerun Startup, Rerun Last Command, Open Shell Only, or Skip) rather than restoring live OS process state.

Where is my checkpoint data stored?

Locally, under .13x/ in the project: one JSON file per checkpoint and latest.json in .13x/snapshots/, the audit log at .13x/snapshots/audit.jsonl, and per-slot scrollback under .13x/sessions/<slotId>/scrollback.txt. Nothing is uploaded.

Will it run a dangerous command without asking?

No hidden commands run. The Resume Plan previews every command; slots flagged dangerous (rm, git push, deploy, terraform, etc.) or set to Rerun Last Command are unselected by default and require explicit confirmation. The backend re-validates before stopping any running terminals.

What about secrets in my commands or output?

Secrets are redacted twice — in the frontend snapshot build and again server-side — for API_KEY/SECRET/TOKEN/PASSWORD assignments, Bearer tokens and sk-/ghp- keys. 13x does not store environment variables or .env contents in checkpoints, and scrollback persistence is opt-in via the capture toggle.

get started

one project. many shells. zero noise.

13x is local-first and free of telemetry — your code, commands and shells never leave the machine. Join the waitlist for the next build.