Coordinate Claude, Codex, and Gemini agents working on your code in parallel. Composable patterns, cost-aware planning, and a planner that learns from every run.
AI coding agents work alone. For complex tasks, that's not enough.
DAG-based execution coordinates agents through research, implementation, and review phases. Parallel when possible, sequential when needed.
Cost-based optimizer estimates token spend before execution. Adaptive replanner adjusts mid-run. Heuristic learning improves plans over time.
Claude for implementation, Codex for review, Gemini for research. Different model per agent in the same orchestration.
7 patterns you can chain inline. Build custom workflows from proven building blocks.--pattern "tdd | red-team"
Save successful orchestrations as parameterized templates. Run them with different inputs. Build a library of proven workflows.
16 core roles plus 10 focused reviewers: security, performance, API contracts, testing, accessibility, dependencies, and more.
Structured workflows for every type of task.
Research, Implement, Review. The default for most tasks.
Write tests first, implement, verify, review.
Build it, break it, harden it. Adversarial security review.
Two approaches compete. A judge picks the winner.
Form hypotheses, test cheaply, build the winner.
Iterative quality gates with devil's advocate challenges.
Three commands. Works on macOS, Linux, and Windows.
npm install -g swarm-engine && swarm install && swarm doctor
Use from Claude Code, any terminal, or VS Code.
/swarm "add rate limiting"
Agents run in split panes with full team orchestration and live cost tracking.
swarm orchestrate "add auth"
Rich console output with progress bars, phase status, and real-time metrics.
@swarm add rate limiting
Sidebar panel, command palette, and Copilot Chat integration.
Export your agents to work natively in other tools.