Open Source
Claudex
Automated dual-agent code quality orchestrator
About Claudex
HAVEN Intelligence developed Claudex, an open source tool that automates code quality control with two AI agents. The tool demonstrates our approach to AI agent integration: deterministic orchestration where Python owns the control flow and LLMs focus on what they do best.
Claudex runs an automated loop where one agent (Claude Code or Codex) plans and implements changes, while another agent reviews and challenges the result. Python owns the entire control flow, including thresholds, convergence logic, and regression tracking. No quality-gate logic is delegated to the LLMs.
Architecture
Claudex is built on a core principle: LLMs are good at reading and writing code, but poor at making decisions about quality boundaries. That is why Python owns all control logic.
Component 1
Planner agent
The first agent (Claude Code or Codex) receives the task description, analyzes the codebase, and implements changes. The agent focuses exclusively on code generation and has no influence on quality assessment.
Component 2
Reviewer agent
The second agent reviews the result and identifies issues with suggestions for improvement. The reviewer scores the changes, but Python validates whether the score meets the predefined thresholds.
Component 3
Python orchestrator
Python controls the entire loop: convergence checks, regression detection, severity counting, and the decision about when the code is good enough. None of these decisions are left to an LLM.
Key Features
- ■ Dual-agent loop: planner and reviewer run iteratively until convergence
- ■ Deterministic control flow: Python owns all thresholds and quality gates
- ■ Three run modes: tmux for interactive use, dashboard TUI for visual overview, headless for CI/CD pipelines
- ■ Regression tracking: automatic detection of quality degradation between iterations
- ■ Open source on GitHub under MIT license
Source Code
Claudex is open source and available on GitHub. Browse the code, read the documentation, or contribute to the project.
View on GitHubInterested in AI agent orchestration?
Book a free AI Analysis and find out how AI agents can automate processes in your business.
Free AI AnalysisNo commitment. Response within 24 hours.