Orchestrate AI Agents
Across Your Codebase
Autonomous execution with human review. Parallel agents with cross‑run memory, multi-model support, and structured trajectories — you set the scope and review draft PRs.
Features
One Command.
Agent Orchestration.
Point it at your repo, set a timer, and walk away. Agents scout, build, and test — then open draft PRs cycle after cycle. You approve what merges.
Milestone Mode
Batches improvements into clean PRs instead of 50 tiny ones.
Parallel Execution
Runs multiple improvements at once in isolated branches with conflict-aware wave scheduling.
Deep Mode
Finds structural problems, not just lint fixes. Principal-engineer-level architectural review.
Impact Scoring
Proposals ranked by impact x confidence. Quality gating filters out low-value changes automatically.
Adversarial Review
Devil's advocate scoring challenges every proposal before approval — only high-value work gets through.
Cross-Run Learnings
Remembers what failed and what worked across sessions. Avoids repeating mistakes, confirms successful patterns.
Spindle Detection
Catches runaway agents — QA ping-pong, command failure loops, and file churn — before they waste tokens.
Smart Deduplication
Never proposes the same fix twice. Each cycle sees all previous work.
Trust Ladder
Control which categories are allowed each run, from conservative to broad as trust grows.
Formulas
One flag for a security audit, test coverage boost, or type cleanup. Write your own in YAML.
Project Detection
Auto-detects your test runner, framework, linter, and language across 10+ ecosystems for correct CLI syntax.
Guidelines Context
Loads your CLAUDE.md or AGENTS.md into every prompt so agents follow your conventions automatically.
MCP Integrations
Plug in external MCP tools — security scanners, pattern analyzers, linters — that run on schedule and feed results back into the loop.
Put Your Codebase
on Autopilot
Free and open source. Agents run autonomously — you review draft PRs before merging.