toryo
Toryo is an open-source command-line tool that functions as an intelligent orchestrator for multiple AI coding agents. It is designed to manage and improve the output of various AI assistants by automatically delegating tasks based on trust scores, ensuring that only high-quality code changes are retained through a novel quality ratcheting mechanism using git commit and revert operations. The tool integrates with popular AI agents such as Claude Code, Aider, Gemini CLI, and local models via Ollama, allowing developers to leverage different AI strengths for code generation, editing, and review. At its core, toryo employs a trust-based delegation system: it maintains a trust score for each agent based on past performance, assigning tasks preferentially to more reliable agents. When an agent completes a task, toryo evaluates the output; if the change meets a certain quality threshold, it is committed to git, effectively ratcheting up the codebase quality. If the output is unsatisfactory, the change is reverted, and the task is retried using a "Ralph Loop" strategy that can fall back to other agents or prompt for user input. This iterative cycle allows developers to automate and refine code generation workflows, reducing manual oversight while ensuring the codebase continuously improves. The tool is particularly useful in multi-agent development environments where consistency and quality control are critical. By combining version control with agent orchestration, toryo provides a safety net that prevents degraded code from entering the repository. Toryo is written to be extensible and can be integrated into existing development pipelines. Its open-source nature allows customization and community contributions, making it a flexible solution for teams experimenting with agentic workflows in software engineering.
- Category
- AI Agents
- Pricing
- Open Source
Problem solved
Toryo addresses the challenge of managing and quality-controlling outputs from multiple AI coding assistants by automating delegation based on trust scores, automatically reverting low-quality changes via git, and retrying tasks until satisfactory results are achieved.
Use cases
- Automated code generation with multiple AI-driven pair programmers
- Trust-based task delegation across Claude, Aider, Gemini, and local models
- Iterative code improvement with git version control and automatic rollbacks
- Reducing manual review overhead in AI-assisted development
- Experimenting with multi-agent workflows for complex software tasks
- Quality ratcheting to ensure only high-quality code is committed
Audience
- Developers
- ML Engineers
- Hobbyists
- Researchers
Pros
- Ensures only high-quality changes are retained via git-based ratcheting
- Adapts delegation based on agent trustworthiness over time
- Supports multiple AI assistants, both cloud and local
- Automated retries with Ralph Loop for resilient task completion
- Open-source and extensible
Cons
- Requires setup and API keys for integrated AI tools
- Trust scoring may require many iterations to stabilize
- Git-dependent reverts may not handle merge conflicts elegantly
- Limited community and documentation as a new project
- May increase token usage and costs due to retries
Official website: https://github.com/JesseRWeigel/toryo
Canonical LambdaBase page: https://www.lambdabase.com/tools/toryo