Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Sign up
Appearance settings

Autonomous Mode

unohee edited this page Apr 16, 2026 · 1 revision

Autonomous Mode

Autonomous mode allows OpenSwarm to operate without human intervention — picking up Linear issues, running pipelines, and reporting results.


Overview

Cron Schedule (e.g., */15 * * * *)
 → AutonomousRunner heartbeat
 → Fetch Linear issues (Todo / In Progress)
 → DecisionEngine filters & prioritizes
 → PairPipeline executes
 → Report to Discord + update Linear state

Setup

1. Configuration

autonomous:
 enabled: true
 pairMode: true
 schedule: "*/15 * * * *"
 maxAttempts: 3
 maxConcurrentTasks: 4
 allowedProjects:
 - ~/dev/my-project

2. Start the Daemon

openswarm start

Or via Discord:

!auto start */15 * * * * --pair

Pipeline Stages

Default Flow

Draft Analyzer (Haiku) → Worker → Reviewer → [Tester] → [Documenter]
  1. Draft Analyzer: Cheap model (Haiku) pre-analyzes the task — classifies intent, collects Code Registry state, runs impact analysis
  2. Worker: Generates code changes
  3. Reviewer: Evaluates changes — APPROVE, REVISE (with feedback), or REJECT
  4. Worker ↔ Reviewer Loop: Iterates until approved or max iterations reached
  5. Tester (optional): Runs test suite
  6. Documenter (optional): Updates documentation

Model Escalation

When a Worker or Reviewer fails repeatedly:

worker:
 model: claude-sonnet-4-20250514
 escalateModel: claude-opus-4-6
 escalateAfterIteration: 3

After 3 failed iterations, the adapter automatically switches to the escalation model.


Pace Control

OpenSwarm enforces rate limits to prevent runaway execution:

  • 5-hour rolling window: Maximum tasks within any 5-hour period
  • Per-project limits: Configurable max tasks per project
  • Exponential backoff: On failures, wait time increases exponentially
  • Turbo mode: Override limits temporarily (4h auto-expire)

Turbo Mode

!turbo on # 5-min heartbeat, 20 daily cap, 4h auto-expire
!turbo off
!turbo status

Task Decomposition

For complex tasks, the Planner agent decomposes them into subtasks:

decomposition:
 enabled: true
 thresholdMinutes: 30
 plannerModel: claude-sonnet-4-20250514

If the estimated task duration exceeds thresholdMinutes, the Planner breaks it into smaller, independently executable subtasks.


Approval Flow

By default, tasks require manual approval:

!approve # Approve the pending task
!reject # Reject the pending task

Enable auto-approval:

!auto approve on

Discord Commands

Command Description
!auto Show status
!auto start [cron] [--pair] Start autonomous mode
!auto stop Stop
!auto run Trigger immediate heartbeat
!auto approve on/off Toggle auto-approval

Monitoring

Discord

The bot reports to the configured Discord channel:

  • Task start/completion/failure
  • Reviewer feedback
  • CI status changes
  • Stuck detection alerts

Web Dashboard

openswarm dash

Real-time view of: pipeline stages, cost tracking, worktree status, and live logs.

Logs

npm run service:logs # stdout
npm run service:errors # stderr

PR Auto-Improvement

When enabled, OpenSwarm monitors open PRs and automatically:

  • Fixes CI failures
  • Resolves merge conflicts
  • Retries until all checks pass
prProcessor:
 enabled: true
 schedule: "*/15 * * * *"
 cooldownHours: 6
 maxIterations: 3

Clone this wiki locally

AltStyle によって変換されたページ (->オリジナル) /