Omnigrex brings specialized AI agents together around your software project. They can shape an idea, plan a change, write code, review it, and test it - while you steer important decisions on GitHub and the project keeps what the team learns.
Omnigrex is about the team around the code: who takes on a problem, how they work together, and what remains after the work is done.
Give agents different specialties, from product and architecture to development, review, and QA. A small fix and a large feature need different teams.
Requirements, plans, decisions, reviews, and lessons belong in Issues and the repository, so the next piece of work can build on them.
Agents handle the everyday back-and-forth. You set direction, answer the questions that matter, and see the results where your team already works.
Can we create an autonomous organization of agents that, across many consecutive tasks, can maintain product, architecture, and project-knowledge consistency without constant human supervision?
The project's core hypothesis - project concept
Start with a GitHub Issue, even a short one. The team can refine the requirements, work out a plan, build the change, and check the result together.
This is one possible path. A quick fix can go straight to coding and review; a larger feature can involve more specialists or be split into smaller Issues.
Choose specialties, responsibilities, and models, then give the agents a shared understanding of your codebase and conventions.
The agents gather context, clarify requirements, and decide how much planning the work needs.
Developers, reviewers, and testers work from their own perspectives, consult each other, and produce a Pull Request and useful project artifacts.
Answer a product question, resolve a blocker, or decide whether a reviewed change belongs in your project.
Omnigrex keeps the process predictable while the agents decide how to tackle the software-development work.
A Go orchestrator routes work between specialists, keeps it moving, and brings questions to a human when needed. PostgreSQL helps it recover after interruptions.
Instructions and decisions live with the code; Issues and Pull Requests carry the conversation. The next agent can pick up the context without relying on someone else's chat history.
Each specialist can keep its own context for a piece of work. OpenCode runs in isolated Docker containers, and conversations can continue after a container is replaced.
Agents read and act through controlled tools while GitHub remains the human-facing home for issues, reviews, decisions, and progress.
Give different agents different responsibilities. Let them consult each other when a problem crosses the line between product, design, code, and quality.
Turn a rough request into clear requirements and a technical approach, consulting each other or you when a decision needs a different perspective.
Works from the requirements and project context, changes code, adds tests, and opens a Pull Request.
Check the change from different angles: code quality, project goals, and how the feature works for a user. Send problems back for another pass.
Shape the team's specialties and autonomy, answer questions that need human judgment, and decide what belongs in your project.
How Omnigrex keeps autonomous work understandable and useful.
Start work, follow progress, and make decisions through GitHub, alongside the Issues and Pull Requests you already use.
Project context lives in Issues and the repository, so changing AI models does not mean starting from scratch.
Each agent has its own responsibilities, permissions, and perspective. A review is more than asking the coding agent to check its own work.
A small fix needs a lighter path than a large feature. Bring in the specialists and planning the problem calls for.
Save important product and architecture decisions with their rationale, so future work can understand why they were made.
Bound reviews, retries, and back-and-forth. When progress stalls or judgment is needed, bring a clear question to a human.
Omnigrex is already helping build itself. Its Developer and Reviewer have worked on multiple merged Pull Requests in this repository, from agent attribution to published deployment images.
Read the code, browse the agent-built Pull Requests, or try the published images and deployment files with the operator guide.
Bring your ideas, feedback, and contributions to the GitHub repository. The team and the project can grow together.