CodeHerder
Command your fleet of coding agents.
What problem does it solve?#
Running many AI coding agents at once creates blind spots. You can't see what each agent does. Agents edit the same files. Cost spreads across machines and repos.
Who is it for?#
Engineering teams and studios that run fleets of AI coding agents on many machines and repositories.
Why did we build it?#
We run AI coding agents on every client engagement, often several at once across different repos. Early on, that meant a wall of terminal tabs. Nobody had a shared view of what each agent did. We lost more than one afternoon to two agents that edited the same file.
Cost became a problem too. Agent time is not free. When a fleet runs on many machines, the spend scatters like the visibility does. Nothing showed what a task cost until the invoice arrived.
We built CodeHerder to fix our own operation first. It gives us one place to see every agent on every machine and repo. Task claims cannot collide. It tracks cost per task, so we do not rebuild it later.
How does it work?#
- A shared task registry coordinates every agent
Agents do not need to share a filesystem or a process. They claim tasks in a central registry instead. The registry never gives two agents the same file at the same time, even on different machines.
- Collision detection at the task level
Before an agent starts, CodeHerder checks what else is in flight on that repo and file set. It rejects a conflicting claim at the start. You do not find it later as a merge conflict or an overwritten edit.
- Per-task cost ledger
Every agent run reports its own spend to its task. Cost then rolls up by task, repo, or machine. You do not get one provider bill with no detail.
- One dashboard across every repo and machine
Every agent reports to the same registry, on any machine. The fleet view is a live query against that registry.
What's the hard part?#
Collision-free tasking without shared state was the hard part. Agents run as independent processes, often on different machines. They have no lock file or shared memory. So a central claim-and-check step must enforce one rule: two agents never touch the same file. That step must be fast, so it does not slow an agent down. It must also be consistent. A claim made on one machine must be visible to an agent that starts on another.
What does it do?#
- Real-time visibility into every agent across machines and repos
- Task claims that stop agents from overwriting each other's work
- Per-task cost tracking from a single dashboard
- Coordinate agents across many repositories at once
Where can you see it work?#
CodeHerder runs the agent fleet that builds this website. Each change goes through plan, code, review, merge and verify.
This website is our live case study.
Our agent fleet builds and maintains customlabs.io. Each change is one task on its own branch. It must pass review and verify before it reaches main. We publish the numbers. We also publish every defect that got past the agents.
- tasks merged to main
- 203
- days
- 109
- defects published
- 12
- build checks added after them
- 13
Counts as of 2026-10-21, at commit 09d8993.
What does this prove we can do for you?#
We build the same discipline for clients who run their own agent fleets or multi-step workflows in production. It has three parts: a central coordination layer, collision detection, and cost accounting per unit of work.
Where this shows up in client work#
Questions, answered.#
The questions we get asked most, with plain answers.
01 What is CodeHerder for?
CodeHerder coordinates many AI coding agents that run on different machines and repositories at once. You see what every agent does. Agents do not collide on the same files. You track what each task costs from one dashboard.
Link to this answer: What is CodeHerder for?02 Does it work with any AI coding agent?
CodeHerder coordinates at the task and repo level. It sits alongside whichever coding agents you already run.
Link to this answer: Does it work with any AI coding agent?03 Do I need to change how my agents work to use it?
No. You do not rewrite your agents. They claim tasks in the CodeHerder registry. The coordination happens outside their process.
Link to this answer: Do I need to change how my agents work to use it?A coordination platform for many AI coding agents across machines and repos. It gives live visibility, collision-free tasking, and cost tracking per task from one dashboard.
This is the same discipline, real architecture decisions and an honest account of the hard part, that we bring to client engagements.