moai-adk
Agentic development harness for Claude Code — SPEC-driven plan/run/sync, TRUST 5 quality gates, model+effort routing, and Claude×GLM multi-LLM cost control. Sin
What is moai-adk?
Agentic development harness for Claude Code — SPEC-driven plan/run/sync, TRUST 5 quality gates, model+effort routing, and Claude×GLM multi-LLM cost control. Single Go binary, 16 languages, zero deps.
Overview of moai-adk
What it is
MoAI-ADK is a verification-driven agent orchestration harness designed to make code produced by Claude Code more trustworthy. It enforces structure from the outside, addressing the limitations of stochastic language models that cannot remember context across turns or assess result quality on their own. The project is a single Go binary with zero dependencies and supports 16 languages.
Key features
- SPEC-driven plan/run/sync workflow for structured agentic development.
- TRUST 5 quality gates to enforce code trustworthiness.
- Model and effort routing for cost control across different AI backends.
- Multi-LLM support including Claude, GLM, and Codex models.
- Factory Mode (v3.2) that splits work across a leader session and multiple numbered lane sessions for parallel execution.
- Backlog queue with
/moai gtdtask management commands, backed by SQLite. - Automatic lane numbering and run recovery for fault tolerance.
Use cases
MoAI-ADK is suited for developers using Claude Code who need to manage complex, multi-step agentic coding tasks with reliability and cost control. It is particularly useful for projects requiring structured planning, execution, and review across multiple parallel work streams, such as large codebase refactoring or multi-feature development. The factory mode enables efficient use of context windows by isolating work per lane.
Getting started
bashmoai cc -f # leader — opens the factorymoai cc -l # a lane, each in its own terminal — joins as the next free numbermoai cc -l # lane-1, lane-2 … in ordermoai glm -l # a lane on the GLM backendmoai codex -l # a Codex lane
License and pricing
The project is licensed under Apache-2.0 and is open source.
Based on the project's public README and repository metadata.