You open three terminal tabs. One runs Claude Code on a bug fix, the second on a documentation pass, the third on a refactor you started yesterday and half-forgot. Twenty minutes later you have lost the thread — which tab finished, which is blocked on your input, which one quietly errored out three commands ago. Running coding agents in parallel is genuinely useful. Running them across a row of identical terminal tabs is not.
Agetor is one answer to that gap. It is an open-source orchestrator that puts a Kanban board in front of your coding agents, so each task becomes a card you can see and move instead of a tab you have to remember. The project sits at version 0.0.1, which makes this a first look at an idea rather than a verdict on a finished tool. The idea is worth understanding anyway, because the problem it targets is real and grows harder as agents get more capable.
What Agetor does at 0.0.1
Agetor describes itself as a harness orchestrator. A harness, in this context, is the program that wraps a coding agent and gives it somewhere to run — Claude Code is a harness. Agetor sits one level above that. It manages multiple harness sessions and represents each one as a card on a Kanban board.
The workflow is straightforward. You create a task, hand it to an agent, and watch the card move through columns — backlog, in progress, review, done, or whatever set of states matches how you work. Instead of context-switching between terminals, you read a board and see the status of every agent in one glance. A card waiting on your review looks different from one still working, and that distinction — what needs me versus what is fine on its own — is the hardest thing to track when you fan work across tabs.
The headline integration is Claude Code. Agetor is built to drive Claude Code sessions as the agent behind each card. Codex support appears on the roadmap but has not shipped; the project lists it as planned. For now, treat Agetor as a Claude Code companion specifically, not a universal front end.
Because the project is open source, the orchestration layer is not a black box. You can read how Agetor starts a session, how it tracks state, and how it decides a task is finished — which matters when you are weighing whether to trust a tool with several agents running while you look away.
"Harness orchestrator" is a term worth keeping straight. The harness is the agent runner — Claude Code itself. The orchestrator is the layer that runs several harnesses and coordinates them. Agetor is the second thing. It does not replace Claude Code; it conducts copies of it.
Where it fits in the agent-management ecosystem
Agetor is not landing in empty space. Over the past year a small category has formed around one observation: a single developer can now supervise several agents at once, and the terminal is a poor surface for that. The tools in this category take different shapes. Some are git-worktree managers that isolate each agent on its own branch. Some are dashboards that stream agent output into a single window. Some are task queues that feed work to agents in sequence.
Agetor's bet is that Kanban is the right mental model. That is a defensible choice. Developers already think in boards, and a board maps cleanly onto the real situation — several units of work sitting in different states at once. The risk is ceremony. If creating a card and dragging it between columns is slower than typing a prompt into a terminal, the abstraction costs more than it returns. Whether Agetor succeeds will come down to how light that interaction stays as the project adds features.
What you should not expect at 0.0.1 is stability, a plugin ecosystem, or visual polish. A version number below 0.1 is a signal in itself: interfaces will change, features are partial, and you will meet edges that a later release sands down. Judge it as a proof of concept, because that is what the version number says it is.
Agetor is at release 0.0.1. Do not point it at a production repository or leave it running unattended on work you cannot afford to lose. Use a scratch repo or a branch you are willing to throw away until the project reaches a stable release.
Should you try it yet?
It depends on what you want from it. If you are looking for a dependable daily driver, 0.0.1 is too early — you would be testing the tool more than using it, and filing issues more than shipping code. If you are curious about agent orchestration as a category, or you already run several Claude Code sessions and feel the tab-juggling cost every day, Agetor is worth cloning and running against a throwaway project for an afternoon. Early open-source releases also depend on real bug reports, so kicking the tires now is a contribution, not just a trial.
If you need agent-driven coding work to happen today and cannot wait for the orchestration category to settle, the pragmatic move is to lean on a tool that is already stable and keep Agetor on your watch list for later.
The longer-term question is whether orchestration belongs in a separate tool at all, or whether harnesses like Claude Code will grow their own multi-task views and absorb the job. Agetor is a bet that the orchestration layer deserves to be its own project — open source, eventually harness-agnostic, and focused on the board rather than the agent. At 0.0.1 that bet is unproven. It is also a clear one, and the problem it names is not going away.
Originally published at pickuma.com. Subscribe to the RSS or follow @pickuma.bsky.social for new reviews.
Top comments (0)