OpenAgents — AI agent framework for plan-first development workflows
OpenAgents is an open-source framework that structures AI-driven software development around a "plan-first, approval-based" workflow. Designed to be integrated with the OpenCode CLI, it helps teams and individual developers use autonomous agents to propose implementation plans, request approval, execute changes incrementally, and run built-in quality checks (tests, type checks, code review) as part of the flow.
Key concepts and features
- Plan-first workflow: agents propose a step-by-step implementation plan before making changes; the human approves each plan (or step) to keep control.
- Approval-based execution: actions are performed only after explicit approval, reducing unsafe or unwanted automated changes.
- Incremental execution & validation: implementation is applied in small steps with automatic validation (tests, linters, type checks) between steps.
- Multi-language support: templates, subagents, and tooling work across TypeScript, Python, Go, and Rust (and can be adapted to other languages).
- Built-in quality tools: automated testing, reviewer subagents for code review and security checks, and validators to ensure consistency with project patterns.
- System Builder: an interactive generator that conducts a domain interview and outputs a tailored agent architecture, subagents, commands, and context files for a given domain.
Architecture
OpenAgents organizes functionality into main user-facing agents (e.g., openagent, opencoder, system-builder) and specialized subagents (task-manager, coder-agent, reviewer, tester, build-agent, image-specialist, domain-analyzer, etc.). The main agent coordinates the workflow, delegates to specialists, and enforces plan + approval stages. Context files (.opencode/context/) encode coding patterns and project-specific standards so agents can follow the project's conventions.
Typical workflow
- User issues a request or task to an agent (for example: "Create a React todo app with TypeScript").
- The agent analyzes the request and proposes a multi-step plan.
- The user reviews and approves the plan (or modifies it).
- The agent executes steps incrementally, running tests and validators after each step.
- Specialist subagents are invoked as needed (tester, reviewer, build-agent).
- The agent summarizes the results and confirms completion.
Installation & profiles
OpenAgents provides multiple installation profiles (Essential, Developer, Business, Full, Advanced) and an interactive installer. The Advanced profile includes the System Builder for generating complete AI systems. The repo contains scripts for both interactive and non-interactive installs and supports macOS, Linux, and Windows (via Git Bash / WSL).
Use cases
- Accelerating routine coding tasks while keeping human oversight.
- Automating multi-step development workflows (feature implementation, refactors, tests).
- Generating domain-specific agent systems using the System Builder.
- Embedding project-specific coding patterns so agents follow team conventions.
Safety and governance
OpenAgents emphasizes safety via the approval-first model, configurable permissions, secure handling of credentials through environment variables, and input sanitization. The framework also provides options for collision detection and safe updates when installing or updating components.
Who maintains it
The repository is maintained by the GitHub user/owner Darren Hinde (darrenhinde). The repo README documents contribution guidelines, tests, and best practices for extending the system.
Where it fits
OpenAgents is suitable for teams and developers who want to integrate AI-assisted development into existing workflows with strong human-in-the-loop controls and built-in quality assurance. Its multi-language and modular design make it adaptable to many codebases and automation scenarios.
