Zozo vs Herdr, cmux, and Conductor
Herdr, cmux, and Conductor give you a better place to run a coding agent: panes that survive a closed lid, a sidebar that shows which agent is blocked, a diff viewer with a merge button. They're good at that. Every one of them still needs you to decide what runs next, type the prompt, answer in the pane, and open the pull request.
I built Zozo because that's the part I kept getting stuck being. Zozo watches the GitHub repositories you point it at, picks the next Issue you've labelled ready, works it in its own worktree on a machine that stays up, asks you once if the spec is missing a decision, and opens a Pull Request that closes the Issue. You label in the evening and review in the morning.
Side by side
Equivalent tiers where they exist. Where a capability needs a paid plan, an integration, or a script you'd write yourself, the cell says so.
ready-for-agent and it's picked in priority order when a slot is free, after its blockers close. You never press start.Competitor cells come from each product's own site, docs, README, or pricing page, read on 2026-09-07: Herdr 0.9.0 and its docs, the cmux README and site, Conductor's docs, pricing page, and changelog through 0.84.2. Tell support@zozohq.com if a cell is out of date and it'll be corrected.
Selection
Say it's Thursday evening and three client repositories each have a couple of small Issues you've been meaning to get to. In Herdr or cmux you'd open three panes and type three prompts, and the fourth Issue waits until you're back. In Conductor you'd create a workspace from each Issue, one at a time.
With Zozo you label the six Issues ready-for-agent and close the laptop. Auto-dispatch picks them in priority order, one Issue per free slot. When a Run finishes, the freed slot picks the next Issue immediately instead of waiting for a timer. An Issue marked blocked on another one waits; a reconciler rechecks every open Issue on every watched repository once a minute, so the moment its blocker closes it's eligible again.
It also knows when to stop. Three attempts in a row failing for the same reason trip a circuit breaker and end the pass. If GitHub can't be read for one repository, that repository is skipped and the others keep moving.
ready-for-agent across three Projects. #214 is blocked on #209.ready-for-agent.Clarification
Every agent can ask a question. The difference is what happens when you're not sitting in front of it. In Herdr the pane goes blocked and holds the session open until you come back to that pane. In Conductor the question sits in the workspace chat. Either way the answer has to be typed where the agent is running, and while you're away the work sits.
Zozo treats a missing decision as a state, not a prompt. When an Issue turns out to need something that isn't in the Issue, the docs, or the codebase, the attempt pauses, labels the Issue awaiting-answers, posts its questions as a comment on it, and sends a push notification to your phone. There's no timeout. You answer from the Dashboard on whatever device you have, the same Thread resumes on the same worktree with every file it already touched, and a digest of the exchange lands on the Issue so the next person reading it doesn't have to ask again.
For a contractor that's the difference between a client's vague ticket costing you a twenty-minute session and costing you one sentence.
Landing
Conductor gets you to a merge button with a Checks tab beside it, which is a real improvement over a raw terminal. Herdr and cmux stop at the pane; pushing and opening the pull request is on you or the agent. In all three, merging is a thing you do.
Zozo opens the Pull Request itself, with a closing reference to the Issue. What happens next is a per-Project setting. Open-PR-only stops there and waits for your review. Auto-merge sends it through Landing, which merges only if the base branch the checks ran against is still an ancestor of the current base. If main moved, it updates and re-verifies instead of merging stale. If checks fail or never report, it stops with a landing-blocked label and a push notification rather than merging anyway. Merging past failing checks is a separate, per-Project switch that is off by default.
Landings run one at a time, so two overnight Pull Requests can't interleave into a broken main. A successful Landing closes its Issue and comments with the branch it landed on, which is what lets the next blocked Issue move.
Closes #214. Enters the Landing Queue behind PR #231.landing-blocked. Push sent. Not merged.Credentials
In a terminal, an agent runs on whatever account the CLI is logged into. If you do client work and your own work on the same machine, keeping a client's usage off your personal subscription is a matter of remembering to switch. Conductor lets you bring your own keys, which helps, but the accounting is still yours to keep.
Zozo makes the billing identity part of the Project. Adding a repository asks you to name a Credential for dispatch and one for intake, with no global default that could silently fill it in. A client's repositories run on an API key on the client's own vendor account, metered by Anthropic or OpenAI with no markup from Zozo. Your own repositories run on the Claude or Codex subscription you already pay for, which adds no inference cost at all. Every Run reports its tokens and cost, and the Dashboard rolls them up by day, Provider, and Credential.
When a client asks what last month's agent work cost, the number is measured on their account, not estimated from yours.
Where Zozo won't fit
These are the reasons a reasonable person on this page picks something else. None of them are hidden in the pricing.
You don't get a pane to type into. Interactive pairing, spiking a design with an agent, or watching output scroll by in real time is what the other three are for. Zozo is for work you've already decided to hand off.
Zozo drives Claude Code, Codex, and OpenCode. If your workflow runs on Cursor, Copilot CLI, or another agent, or your issues live in Linear rather than GitHub, it doesn't fit today.
Zozo runs on a VM we operate for you. There's no self-hosted build yet and no date promised for one. Signup is by invite code or the waitlist until open signup passes its readiness checks.
Next
The core loop is idea to Issues to code to Pull Request, and every step of it is on the home page. The trial is fourteen days on Standard, with your own Claude or Codex subscription and nothing added on top of it.