AI Coding Agent Orchestrator for Team-Controlled Execution
Use Mrrlin as an AI coding agent orchestrator to coordinate Codex, Claude Code, worktrees, reviews, and evidence across engineering tasks.
Chat tools vs Mrrlin
The execution layer beats another blank chat window.
Chat tools make you re-send context and manage the work. Mrrlin keeps project memory and spends tokens deliberately.
Your agents, under one harness
One coding agent gives you output. Mrrlin gives you a harness.
Codex, Claude Code, and other agents can write. Mrrlin keeps the brief, routes each task, challenges weak work, and brings you in only for decisions that need you.
Read the workflow transcript
Without a harness, a person carries context between prompts, checks each output, follows up, and decides what happens next. With Mrrlin, one goal becomes a visible plan: work is routed, reviewed, retried when weak, and brought back only for the decisions that need a human.
Why it matters
Someone already orchestrates your coding agents. Usually it is you.
Mrrlin gives coding agents a durable task, project memory, review gates, and a visible execution trail so teams can move faster without losing control.
Workflow
The whole loop an orchestrator has to own, from plan to memory.
Plan the goal into tasks
From one goal, the Director drafts a plan of tasks with dependencies, acceptance checks and an impact estimate for each. Important calls go to a spec review in which architect, security and scope reviewers, each on its own model, must converge before implementation starts.
Route each task deliberately
Routine work goes to cheaper models, and repository-heavy work goes to Claude Code or Codex CLI running locally on your subscriptions. Settings is where a kind of work gets pinned to one model or left on auto, and reviewers are drawn from a provider other than the executor's.
Isolate and record every run
Each run executes in its own git worktree on its own branch, and its record shows status, base commit and provider. Plans, checkpoints, test results, errors and diff summaries pile up as artifacts, so the run explains itself after the terminal closes.
Review, receipt, remember
A separate reviewer reads the result and sends weak work back with notes. Closing the task takes proof such as a passing build or a working URL, and what was decided, and why, is written back so the next task starts warm.
Use cases
Where it fits.
Feature implementation
Take a landing-page improvement from idea to review. The Director scopes it into copy, component and tracking tasks, each agent works on its own branch, checks run, review notes accumulate, and you receive a pull request with its evidence already on it instead of a branch to investigate.
Bug fix triage
Paste fifteen bug reports into the Director. Each becomes a task with reproduction steps, an impact level and a check that proves the fix. Agents take the reproducible ones, while reports nobody can reproduce become inbox questions or close with no result instead of lingering.
Repository cleanup
Removing the dead code of an abandoned onboarding experiment is tedious and risky in one sweep. Split it into small tasks per module, each with its own worktree, pull request and recorded tests, so a reviewer approves one safe deletion at a time.
Code review preparation
Before a person opens the pull request, it already carries a diff summary, test results, the executor's self-review and notes from a reviewer on another model. The human reviewer starts from intent and evidence, and spends the time on judgment rather than reconstruction.
Comparison
Coordinating agents by hand versus an orchestrator that owns the loop.
Coding agents
Every guide in this section.
Keep reading
More on coordinating coding agents.
FAQ
Before you start.
How is an AI coding agent orchestrator different from a harness or a dashboard?
A harness is the containment around each run: context, isolation, review and approval. A dashboard is the view of what is happening. An orchestrator decides what runs, on which agent, in what order, and what happens after a result comes back. Mrrlin covers all three, because routing work without containing it or showing it only moves the chaos somewhere else.
Which coding agents can Mrrlin orchestrate?
Mrrlin works as an AI coding agent orchestrator for Claude Code and Codex CLI, which run locally through its bridge on subscriptions you already have; GPT-5 is available too, through its own local CLI. Cheaper models take routine tasks, while Gemini and others act as independent reviewers. You can drive it from Claude Code, Codex CLI, Cursor, the web app or a phone.
Do we have to move our repositories or CI to adopt an AI coding agent orchestrator?
No. Mrrlin reaches your repository through a grant that covers branches, pull requests and workflows, with every call logged. Deploys run through the GitHub Actions workflows you already have, for preview and production, and production waits for approval. The agents themselves execute on your own machine rather than on infrastructure you would need to adopt.
Orchestrate Claude Code & Codex CLI without terminal lockups.
Stop juggling disconnected CLI windows and resolving manual merge collisions. Mrrlin coordinates your agents in isolated git worktrees, collects passing test artifacts, and opens verified pull requests.
No credit card · No migration · One goal