Multi-AI workflow orchestrator with an MCP coordination server.
You architect. Agents build and review. MySwat connects them.
MySwat automates the copy-paste routing between AI agents. It persists shared workflow state to TiDB, exposes coordination through an MCP server, and loops dev/QA review cycles until LGTM — while you stay in the architect seat.
New to MySwat? Follow the Installation Guide for step-by-step setup: cloning, TiDB Cloud configuration, agent CLI installation, and running your first workflow.
# Start the daemon
myswat server
# Initialize a project (in another shell)
myswat init "my-project" --repo /path/to/repo
# Queue a task
myswat work -p my-project "Implement bloom filter for compaction"
# Follow progress live
myswat work -p my-project "Add caching layer" --follow
# Interactive chat
myswat chat -p my-project
# Inspect state
myswat status -p my-project --details
myswat search "bloom filter" -p my-project You (architect)
|
| "Implement bloom filter for compaction"
v
MySwat daemon
| 1. Queues stage assignments in TiDB
| 2. Starts managed workers for each agent role
| 3. Workers claim work through MCP, agents execute
| 4. WorkflowKernel advances or loops review stages
| 5. Final report + persisted team knowledge
v
Done
| Mode | CLI flags | What runs |
|---|---|---|
| Full | (default) | Design, review, plan, develop, QA test, report |
| Design | --design |
Design + plan with reviews, no code |
| Develop | --develop |
Phased implementation with QA review |
| Test | --test |
GA test plan, execute, bug fixes, report |
- Python 3.12+
- TiDB Cloud account (free tier works)
- At least one AI CLI: Codex, Claude Code, or Kimi
- Installation Guide — full setup walkthrough for new users
- CLI Reference — all commands, work modes, chat commands
- Configuration — config file, environment variables, per-backend setup
- Architecture — components, memory tiers, knowledge pipeline, TiDB schema
MIT
