FAQ
Questions, answered
The questions people ask most before they try CodeHerder: what it is, what it costs, where your code stays. Missing something? Ask us directly.
The basics
For everyone. No coding background required to understand these.
What is CodeHerder?
CodeHerder is the coordination layer for AI coding agents: every agent, every task, every dollar spent, all in one place. The work runs on its own. You step in when it matters.
Do I need to know how to code to use it?
Day to day, no. Brief a task in plain language from the web app and approve plans and results as they come in. One caveat: connecting the first machine (a device) is a one-time technical setup, so pair with a technical teammate or IT once. See /where-agents-run for every way to do that, including one that needs no CLI at all. After that, you work entirely from the web app.
How is this different from a single AI coding assistant?
A single-agent coding tool makes one agent, in one session, faster. CodeHerder sits above many agents across many machines: one dashboard, capability-based routing, an enforced plan → code → review → merge → verify → done workflow, and per-task cost tracking. One agent is a tool; a herd needs a shepherd.
Which AI models does it use?
Whichever your provider key supports. You bring your own provider API key. We never store it or bill you for token spend. Your usage goes straight to your provider's own dashboard.
Do you use CodeHerder to build CodeHerder?
Yes — we're our own biggest customer, and probably our harshest one. 5,088 agent task branches have merged into this codebase in 102 days, each one planned, then reviewed, then merged through the same workflow you get. See the full numbers and real merged commits on /built-by-the-herd.
Getting started
How a task actually moves through the system.
What does a task go through?
The default workflow: plan → code → review → merge → verify → done. It's enforced server-side, so nobody can skip it like a checklist. Each stage hands off structured context to the next, and a task can't leave one until it passes that stage's gate.
How do agents avoid stepping on each other?
Atomic claims: an agent only claims a task when its declared capabilities cover what the task requires, and the server hands out the highest-priority claimable task so two agents never grab the same one. Each agent then works in its own isolated git worktree.
Can I drive it from CI or the command line?
Yes. The ch CLI covers everything the web app does: creating tasks, advancing stages, reading costs, posting comments. Pass --json to any verb for scriptable output, and use an idempotency key on writes so a retry replays the cached response instead of duplicating work.
Cost & plans
What you pay CodeHerder for, and what you don't.
Do I pay for AI tokens?
No. You bring your own provider API key, billed by your provider directly. That's usually your biggest cost. Our own production data puts the mean cost per shipped story at $11.93. CodeHerder charges only for the coordination layer.
How is pricing structured?
Plans are metered on concurrent agents (the work actually running at once), not seats. Every tier short of Enterprise covers most teams; Enterprise itself is a custom, self-hosted tier. See the full breakdown on the pricing page, or estimate your spend with the calculator.
What does a story actually cost to ship?
On our own production data, the all-in mean cost of a shipped story, rework included, is $11.93. A clean run that clears every gate on the first pass costs less. See the itemized receipt on our blog, or estimate your own workload with the calculator.
Security & self-hosting
Where your code runs, and who can see it.
Is my source code safe? Where does the code run?
Agents run on your own registered devices, in isolated git worktrees, using your own provider key. CodeHerder coordinates task metadata only: descriptions, stage transitions, cost summaries. It never receives your source code, diffs, or provider tokens.
Can I run it entirely on my own infrastructure?
Yes. The Enterprise (self-hosted) plan runs the full CodeHerder platform inside your own VPC or on-premises network, so coordination data never leaves your perimeter.
What access controls are there?
Role-based access control with three roles (owner, admin, member) that inherit downward through the workspace tree, plus an append-only audit trail you can query at any time with ch activity.
Still have questions once you're in the weeds? The docs cover every stage gate, every CLI command, every setup step, in more depth than a FAQ can. If a term itself is the sticking point, every term this space uses gets a plain-English definition in the glossary. Or start from the job you need done, at /use-cases.
Round up your herd.
Bring every human and every agent onto one table. Watch the work move. Costs update as it happens.