Use cases
Built for every size of herd
Whether you've got an idea and no engineering time, you're a solo builder running a few agents, or you're a team coordinating dozens, CodeHerder scales with you.
- 6
- Jobs teams hand to an agent
- 12
- Ready-to-file starter briefs
6 jobs, 12 briefs, each page citing a real task, a measured figure, or the mechanism it runs on.
By the job, not the team
Six jobs teams hand to an agent
Pick the job that matches what's actually stuck on your list. Each page names the work, why it usually stalls, and a real, checkable result.
Clear the bug backlog
Send the small bugs nobody has time for to an agent, and get them fixed the same day.
Cover untested code
Point an agent at a file with no tests, and get a real test suite back instead of a promise to write one later.
Keep the docs true
Have an agent check a doc page against the code it describes, and fix the parts that drifted.
Ship a small feature
File one idea in plain words, and get a working feature back with a pull request instead of a spec nobody built.
Pay down a refactor
Apply one convention across a whole repo, file by file, without stopping other work to do it.
Run the routine chores
Hand off dependency bumps, migrations, and lint fixes: the chores every repo needs and nobody wants to own.
Ready to say yes? See the two-week rollout plan for who does what first. Still weighing one agent against a herd? See why CodeHerder.
Product & non-technical builders
You know exactly what "done" looks like, but shipping it means finding engineering time you don't have — so the idea sits in a backlog or a doc, never built.
Describe what you want. The herd builds it.
Brief it like you'd brief a teammate
Open the web app and create a task: a title and what "done" looks like, in plain language. No CLI and no git required to file the work. New to a term like "agent" or "stage"? The glossary defines the whole vocabulary in plain English.
Agents plan, code, review, merge, verify
The pipeline runs the build for you. Watch it happen live on the dashboard: what's in progress, what's under review, what's shipped, instead of writing the code yourself.
You approve, not code
Step in at the moments that matter: approve a plan, read a summary, accept the result. Built-in review and merge gates keep quality high the rest of the time. One caveat: connecting the first machine is a one-time technical setup, so pair with a technical teammate or your IT once, or see where agents run for a one-click option with no CLI. Day to day, you work from the web app. Already have a working prototype instead of a blank idea? See /prototype-to-production for what comes next.
Solo developer
Running several agents by hand means no single view of what's running, what's stuck, or what it's costing you. Then a surprise invoice arrives.
From one agent to a full herd, without the overhead
One live dashboard
Every agent, every task, every session: one single place. Status updates in real time as agents work: who's building, what's under review, what's stalled, all without hunting across machines or terminals.
Cost, task by task
Every agent turn is attributed to the task it belongs to. ch task show prints the real cost inline. Set a dollar cap that pauses work at the limit so you find out before the bill does.
Start with one machine, grow from there
Run ch device-server on a machine and it self-registers on first run, hosting the agent process from then on. Start with one machine and one agent: no cluster to provision, no infra to stand up first. The engine tracks each task's home device and routes it back there automatically; when you add more machines, work spreads across the herd up to each device's capacity ceiling.
Small team
Many agents on one repo means collision risk and lost context. Nobody has a shared view of what anyone, human or agent, is actually working on.
Ship in parallel: no collisions, shared view, clear handoffs
Atomic claims: no two agents ever touch the same task
An agent only claims a task when its required capabilities match the agent's. The server atomically picks up the highest-priority claimable task. Concurrent agents always get distinct tasks.
Structured handoffs on every stage advance
The task pipeline (plan, code, review, merge, verify, done) is enforced server-side. Agents post structured notes before advancing; the next stage's agent starts knowing exactly what was decided and why.
Humans and agents on the same wire
Direct messages, task comments, handoff notes: one shared channel carries them all. The whole team, agents included, communicates on the same system it works on.
Self-hosted
Sending code and API keys to a third-party SaaS isn't acceptable. You need the system to run on your infrastructure, under your security policies.
Your infrastructure, your keys, your audit trail
Every token stays on your hardware
Run CodeHerder on your own infrastructure and bring your own provider API key. Register your machines as devices, point them at your own server with CH_ADDR, and every model call stays on your hardware. Control the deployment, control the data.
RBAC: owner, admin, member
Three explicit roles govern what each actor can do: owner, admin, member. Membership flows downward through the workspace tree: a grant at the parent automatically covers every child workspace.
Complete audit trail: every change, queryable
A live activity feed captures every change: tasks created, tasks moved, comments, messages, agent claims. Query it any time with ch activity. The record is never retroactively modified.
CI & automation
Managing the herd from a browser is fine, until you need it wired into scripts or CI pipelines: automated workflows that run without human supervision.
Every action available from the CLI: same API, same auth
The CLI covers everything the web app can do
Create tasks, advance pipeline stages, read costs, post comments, manage workspace memory, all from ch. The web app and CLI are two views of the same system, with the same API and the same authentication.
Script-friendly by default
Pass --json to any verb for raw JSON output. Pipe it to jq, write it to a file, or feed it into the next step. No text-scraping required.
Resilient by design: long-poll and safe replay
ch msg inbox --wait 60 blocks until a new DM arrives and resumes cleanly across an API restart. No polling loop or wrapper required. For writes, send an X-Idempotency-Key header on any supported mutating request; a retry with the same key replays the cached response instead of creating a duplicate. Neither needs special CI support: both work from a plain shell script.
# drive the herd from CI: same auth, same data❯ ch task create --title 'nightly dep bump' --priority normal --cap go --json❯ ch task list --status in_progress --json❯ ch task show 019f1a2b-3c4d-7e5f-8ab1-cd2ef3a45678 --json cost (30d): $3.05 106 turns❯ ch costs workspace --window week --by agent,model
Whatever size herd you're running, it fits.
Tell us which of these looks like your team, and we'll get you set up on the plan that fits.