CAS coordinates multiple Claude Code agents to plan, build, and test features in parallel on your codebase and your machine.
At a Glance
Pricing
Free during beta. Runs entirely on your machine with optional cloud sync.
Engagement
Available On
Listed Mar 2026
About CAS
CAS (Coding Agent System) is a multi-agent orchestration tool that runs entirely on your local machine, coordinating multiple Claude Code agents to work simultaneously on a single codebase. A supervisor agent breaks down your request into structured tasks with priorities and dependencies, then spawns worker agents that each operate in isolated git worktrees — eliminating merge conflicts and enabling true parallel development. Workers open pull requests when done, keeping you in control of code review and direction. Cloud sync is optional, meaning your code never has to leave your laptop.
- Persistent Memory — Agents remember past sessions, bugs fixed, patterns learned, and your preferences so new sessions start informed, not blank.
- Task Orchestration — Work is broken into prioritized tasks with dependency tracking; agents auto-claim what's ready and skip what's blocked.
- Self-Improving Rules — When you correct an agent's mistake, CAS creates a rule so the same error never happens again.
- Reusable Skills — Teach an agent how to deploy, test, or lint once, and every agent on your project inherits that skill.
- Git Worktrees — Each agent works in its own isolated copy of the repo, preventing file locks, conflicts, and bottlenecks.
- Team Sync — Context learned by one developer's agents can be pushed to the cloud and pulled by the whole team.
- Factory Pattern — A named supervisor coordinates named workers (e.g., keen-panther, wise-lynx) in a visual factory dashboard.
- 100% Local — Install with a single curl command; runs on your machine with no mandatory cloud dependency.
Community Discussions
Be the first to start a conversation about CAS
Share your experience with CAS, ask questions, or help others learn from your insights.
Pricing
Open Source
Free during beta. Runs entirely on your machine with optional cloud sync.
- Multiple Claude Code agents in parallel
- Supervisor + worker factory pattern
- Persistent agent memory
- Git worktree isolation
- Self-improving rules
Capabilities
Key Features
- Multi-agent parallel task execution
- Supervisor + worker agent architecture
- Persistent agent memory across sessions
- Self-improving rules from corrections
- Reusable agent skills
- Git worktree isolation per agent
- Automatic pull request creation
- Task prioritization and dependency tracking
- Optional cloud team sync
- 100% local execution
- Works with Claude Code
