One agent, set up your way
Give a single agent a role and the skills it needs. Use it day to day and keep refining its setup as you learn what works.
Open source meta-harness for coding agents
Stop letting your experience disappear between sessions.Build it into your own agent ecosystem.
An open source playground and meta-harness for developers who want to go beyond a coding agent's default setup. Bring Claude Code, Codex, and others into one ecosystem.
Create agents → build teams → assign a team to a room → give it a task. Open another room with the same team or a different one, give it another task, and repeat.
AgentsCommanderYour agents. Your way.- Create
- Use
- Refine
- Reuse
Start with one agent or a team
Build an agent setup you can reuse
Keep roles, skills, and recorded experience in your own repository. You decide which agents exist and how they work: alone, or grouped into rooms with an orchestrator.
An orchestrator and the agents you choose
Define a team once and open it as a room for each task. The orchestrator coordinates the agents you assigned, each with its own role and skills.
Your repository
Keep what you learn in your repository
Trace how your agents work. Roles, skills, messages, recorded errors, plans, and notes stay as plain text files that you, or a team of agents you’ve set up for that task, can read, review, and version.
This lets you build teams for different tasks and increase your productivity as you gain experience and refine how they work.
- Review the messages and errors your agents leave behind.
- Refine roles, skills, and your own harness from that evidence.
- Reuse the improved setup on the next task.
.ac/lives in your repository_agent_<name>/Role.mdrole you define_agent_<name>/skills/skills you maintain_agent_<name>/memory/notes your agents writeroom-<n>-<team>/messaging/messages they exchangeerrors/errors they recordplans/plans you review
Meta-harness
A harness on top of your coding agents' harnesses
Your coding agents keep their own tools and behavior. AgentsCommander adds the layer around them, so you can refine their setup and automate how they run and coordinate, solo or in teams.
Real terminals, real CLIs
Each agent runs as a local terminal session of the coding-agent CLI you already use, so prompts, failures, and progress stay visible.
Orchestrators that delegate
Give a room orchestrator a goal. It sends markdown briefs to the agents you assigned and collects their replies as messages you can inspect.
Every exchange on disk
Inter-agent messages are timestamped markdown files. Read them, diff them, archive them, or debug them with normal developer tools.
Resume from explicit state
On long work, an agent can write a resume note, clear its own context, and continue from the task state captured in that file.
Compare setups on the same brief
Rerun a task with a different role, skill, or CLI profile, then compare the terminal output, messages, tests, and diffs.
Parallel work in one view
Watch orchestrators and agents side by side, with running, waiting, and exited states, from one desktop window.
Get started
Pick one of two ways. You bring your own CLIs, accounts, and model access.
Community
Coding agents move fast. Let's figure them out together.
Share what you learn. Help build a common way to configure, automate, and use different coding agents from one place.