Maggy
Turns Claude Code, Codex, Kimi and Gemini CLI into a test-enforced, quality-gated workflow with cost-aware routing across model tiers, cross-session memory and a code property graph.
Built for engineering teams standardising how they use coding agents.
- Routes work across model tiers by cost and task complexity
- Enforces tests and quality gates as part of the agent loop
- Keeps memory across sessions, so context survives compaction
- Builds a code property graph for structural queries
- Runs parallel isolated agents, with a plugin system around them
Started as an opinionated Claude Code setup kit and grew into the harness COEUS runs its own engagements through. Published under MIT because the practice is more useful shared than hoarded.
Yours will not be identical.
These are starting points, not shrink-wrap. We fit them to your call flow, your systems and your rules, then quote the work fixed and in writing.