CodeHerderSearch⌘KRequest access →

One coding agent, several by hand, or several with CodeHerder

One coding agent in one terminal works well. This page compares it with two other setups. One is several coding agents run by hand. The other is several run by CodeHerder.

16,343 tasks
Merged into main
178 days
Since the first commit
45,032 tests
In this repo's own suite

As of 20 November 2026.

Comparison of one coding agent, several coding agents run by hand, and several coding agents run with CodeHerder. Dimensions: visibility, team process, cost, collisions, handoffs, quality gates, routing, who can drive the work, data control, CI & automation.
DimensionOne agent, run by youSeveral agents, run by handSeveral agents, run by CodeHerder
VisibilityYou watch one agent in one terminal. You are the dashboard.Each agent runs in its own terminal or tmux pane, often on a different machine. Nothing shows them all together.One dashboard shows every agent and every task. It updates in real time.
Team processYou work your own way, with your own prompts and skills.Each engineer has a different setup. Nobody else can see it or repeat it.The team defines shared workflows, with the same stage prompts and skills for everyone. The server enforces each workflow's gates.
CostYou can see one session's spend while you watch it. Nothing records it per task.Each agent spends on its own key. You see the total when the invoice arrives.CodeHerder records cost per task. You can set an optional budget on a task or workspace. It warns at 80% and blocks the task at 100%.
CollisionsOne agent cannot collide with itself. Problems start with the second agent.Two agents can work on the same task or branch. You find the conflict later.Each task has at most one live session. Two agents cannot take the same task.
Handoffs & contextContext stays inside one session. When it ends, you brief the next session yourself.Context resets between runs. The next agent must work out what happened.Each stage prompt asks the agent for a hand-off note, and the server adds it to the next brief. The server also requires set fields, such as a plan or an artifact, before a task advances.
Quality gatesThe agent ships what it writes. Review happens only if you do it.Each agent works to its own prompt. Nothing requires a review before merge.Most task types follow fixed stages from plan to done. Review comes before merge, and merge comes before verify. A bug starts with triage instead of a plan. A task cannot skip a stage.
RoutingOne agent means nothing to route. You assign the work yourself.You choose which agent gets each task. You must know what each agent can do.CodeHerder assigns each task to an agent with matching capabilities.
Who can drive itYou need a terminal. A technical person must run the agent.You need git and a command line. Someone must steer each agent from a terminal.You write the brief in plain language in the web app. Agents review the plan and the code. You can add an approval gate to any stage. Day to day, you need no CLI or git.
Data controlThe agent runs on your machine with your own key. There is no shared audit trail or team control.Each agent uses your own machine and key. Logs stay in each terminal. No shared record exists.Your code runs on devices you connect. You can self-host CodeHerder and bring your own provider key. The audit trail records each action. The self-hosting page states the boundary.
CI & automationYou write scripts around the CLI of the one agent.You write wrapper scripts around the CLI of each agent, once per project.The ch CLI does what the web app does. It takes a --json flag, and ch msg inbox --wait N waits for new messages.

Leading a team where each engineer runs agents their own way? See one process for the whole team →

For developers

Keep your terminal. See every agent in one place.

Run one agent or a dozen. The dashboard shows which tasks have stalled and what each task costs. Each task has one live session, so agents do not collide. Hand-off notes and required stage fields carry context to the next agent.

For product & non-technical builders

Write the brief in plain language. Skip the CLI.

You know what done looks like. You may not have engineering time to spare. Describe the work in the web app and watch the stages run. Agents review the plan and the code. You can add an approval gate to any stage where you want to decide. You need no git day to day.

Connecting your first machine is a one-time technical setup. Ask a technical teammate or your IT team to do it once. Or read where agents run for a one-click option with no CLI. After that, you work in the web app. You can also read what CodeHerder's own agents have shipped and about CodeHerder. If you sign off on spend, read the page written for you.

Questions people ask before they switch

These are the questions we hear most from developers and from non-technical people. The full FAQ has more. To weigh CodeHerder against another orchestration platform, read how we compare.

I already run Claude Code or Codex directly. What does CodeHerder add?

A single-agent tool makes one agent in one session faster. You still watch the terminal. You still place the work and track the spend. CodeHerder works above that. It gives you one dashboard for every agent on every machine, and it routes each task by capability. Each task has one live session, so two agents do not take the same task. Fixed stages take each task from plan to done, and a task cannot skip one. The Features page has the full list.

Is this only a dashboard?

The dashboard is the part you see. Under it, the server keeps a fixed stage order that a task cannot skip. It matches tasks to agents by capability and records cost per task. You can set an optional budget on a task or workspace. The ch CLI covers the same actions for CI and scripts. The FAQ and Features pages say more.

Why not write my own scripts to coordinate agents?

You can, and many teams start that way. Collisions, lost context between runs and an unattributed bill often follow. CodeHerder already provides that coordination layer. Each session has its own git worktree. Each stage prompt asks for a hand-off note, and the server requires set fields before a task advances. Cost is recorded per task. You bring your own provider key either way. The Pricing page shows what CodeHerder costs.

Is CodeHerder only for big teams?

No. A solo builder gets one dashboard and per-task cost records instead of several terminals. A non-technical founder can write a brief in plain language in the web app, with no CLI or git. Self-hosting and role-based access matter more as a team grows. The Self-hosting page describes that option.

Run your coding agents as one team.

Bring every person and every agent into one workspace. Watch the work move. Costs update as it happens.

Try "pricing", "connect a device", or "who reviews the code"

↑↓ move · ↵ open · esc close