Autonomous Runs & Goals
Autonomous runs let LanMate keep pushing a goal forward without you issuing instructions: set a goal and a budget, and it works round after round, self-checking, until quality gates pass or the budget runs out. Great for “clear direction, long process” work like data cleaning, batch refactoring, and long-document writing.
Starting an Autonomous Run
Three equivalent entry points:
- Composer mode menu: click the mode menu to the left of the input (Agent / Plan / Spec / Goal) and choose “Autonomous Run” — a launch form pops up
- Right panel “Goals” tab: click “Start an autonomous run” in the empty state
- Slash commands:
/autonomous on(with optional budget arguments) and/goal <goal>
For lightweight usage, you can also pick the “Goal” option straight from the Turn mode menu: set the goal in one sentence and the goal loop starts — no budget form. Use this page’s heavyweight mode when you need a budget and quality gates.
Launch Form
| Item | Description |
|---|---|
| Goal | Optional: left blank, the run proceeds purely autonomously, stopping when quality gates pass or the budget tops out; when set, work revolves around the goal |
| Budget tier | Light (default, recommended): up to 3 continuations / 12 rounds / 80k tokens / 30 minutes; Medium: 10 continuations / 30 rounds / 500k tokens / 2 hours; Heavy: 30 continuations / 100 rounds / 3M tokens / 6 hours |
| Advanced settings | Expand to fine-tune: max continuations, max rounds, token limit, time limit, quality-gate commands (one per line, e.g., npm test) |
The form reminds you: a running run keeps consuming budget until quality gates pass or a limit is hit; you can stop it at any time.
Controlling a Running Run
- Always-visible status: a persistent “Autonomous Run” status line above the composer (round count, token usage, whether a limit is hit); click it to open the right panel for full metrics
- Adjust budget: “Adjust Budget” in the right panel edits each limit in place — saving takes effect immediately and progress is preserved (equivalent command:
/autonomous limits) - Pause / resume goals: goals can be paused from the goal details area and resumed later
- Stop: one-click stop from the mode menu, the status line, or the right panel (equivalent to
/autonomous off)
Quality Gates
Quality gates are the “acceptance criteria” of an autonomous run: after each round, your commands run automatically (tests, lint scripts, etc.), and all of them must pass for the run to complete. On failure, the right panel shows the failed command, retry count, and latest output, and LanMate keeps fixing accordingly.
End States
An autonomous run ends normally in three ways:
- Quality gates passed — task complete
- Quality-gate retries exhausted — repeated fixes still didn’t pass
- Limit reached — one of continuations / rounds / tokens / time limit hit its cap (the status shows which one)
A notification fires once at the end; the right panel keeps a read-only record of the run for later review.
Safety Boundaries
- No approval exemptions: autonomous runs don’t bypass approval or sandbox policies; high-risk operations still pop approval cards as usual
- Conservative by default: budget starts at the smallest tier; advanced parameters are collapsed by default
- Stoppable anytime: three entry points to stop at any moment
- No fighting: while sub-agents or background tasks are running, autonomous continuations suspend automatically and resume after they settle; when model quota tops out, the run parks automatically and resumes at the reset time
Compared with the “Goal” option in the composer mode menu, an autonomous run is the “heavyweight mode with a budget and quality gates”; the Goal option is a lightweight goal-driven loop for goal focus within a single turn.
Related Pages
- New Task & Conversation: modes and queueing
- Approval and Security: approval tiers