your agents, voluntold.
drop an issue on your board from your phone. an agent picks it up, opens a PR, and merges it on green. your hardware, your account, your code.
the routing layer between your issue board and the AI coding agents on your machines. you assign work once, to a single voluntold identity. it routes the ticket to a connected device and writes the result back.
point voluntold at Linear. it shows up as one assignee, the voluntold identity.
one command on any machine registers it and the agents on it: Claude Code, Codex, whichever you run.
drop a ticket on voluntold. the agent ships it, then gets voluntold the next one.
ask questions against any connected repo. a live agent answers from the device where that repo lives. ask where something lives, why a test broke, or what a change would touch. it reads the real code, not a summary.
a big ticket can become smaller shippable pieces. if an agent hits a bug mid-flight, it files that too. the board fills itself in.
mark a ticket blocked by another and voluntold holds that ticket back. when the blocker closes, it routes the next ticket. decomposed work lands in the right order, however many machines are free.
nowhere new. voluntold routes the work. it never runs the model or holds your code.
Claude Code or Codex, running on your machines. not ours.
your own Anthropic and OpenAI accounts. you run the model. we route the work.
the work runs where you already run it. your code never leaves your machines.
every connection is end-to-end encrypted.
❯isn’t this just another AI coding agent?
no. voluntold doesn’t run an agent at all. it’s the router between your issue board and the agents you already run (Claude Code, Codex) on your own machines and your own accounts. a hosted agent runs in someone else’s cloud. here your code and the model never leave your hardware. voluntold just dispatches the work, keeps it in order, and writes the result back.
❯can I use my own Claude or OpenAI account, or a local model?
yes. it drives Claude Code or Codex, signed into your own Anthropic or OpenAI account (subscription or API key). local and third-party model providers aren’t supported today. it uses the claude and codex binaries you already have.
❯will it burn through my model usage?
you pay for your own usage, so you stay in control. voluntold won’t hand new work to an agent that’s nearly out of quota. it waits for the reset instead of stranding a half-finished ticket.
❯what does it cost?
nothing while we’re in early access. you already pay for your own agents, and voluntold is just the routing layer on top. pricing comes later.
❯does a machine have to stay online? what if my laptop sleeps mid-ticket?
work lives in our database, not on the connection, so machines can drop and reconnect freely. if one sleeps mid-ticket, the work re-queues and the next free machine picks it up. nothing is lost. for hands-off running, keep a machine awake with push credentials. otherwise, just run it while you work.
❯does it merge to main on its own, or do I review first?
the agent opens a PR, runs your CI, and merges on green using your repo’s own merge convention. your CI and branch protection are the real gate. before the PR, it runs an adversarial self-review: a separate reviewer critiques the plan, then the diff. want a human to sign off? turn on branch protection and it stops at the PR.
❯do you see my code, or run the model?
neither. the agent runs on your machine, against your own account. your source never reaches us, and we never touch or pay for the model. voluntold only moves the ticket, its status, and the agent’s narration.
❯which boards does it support, and how does it show up on mine?
Linear and GitHub Issues. it connects as a single @voluntold assignee. assign an issue to it, or @-mention it in a comment, and that comment becomes the instruction.
❯private repos, monorepos, any language?
yes to all. the agent works in your local checkout with your own git credentials, so nothing is ever cloned to us. one caveat today: the open-a-PR, watch-CI, merge-on-green loop uses GitHub’s gh. voluntold is at its best on GitHub right now, with Linear as the board. other hosts are on the way. what matters for results is solid CI and docs, not the language.
❯is it open source? can I self-host it?
voluntold itself is a hosted service, not open source today. but the parts that matter for trust, your code and the model, always run on your machines and your own accounts. never on ours.
❯can I step in, or does it just run off on its own?
you steer from the board. it narrates its thinking and actions on the ticket. when it’s unsure, it asks a question there and parks until you answer. it also takes mid-flight comments or a stop. there’s a live “chat with the repo” when you want to poke around directly.
❯can it handle a big piece of work?
yes. a large ticket is broken into smaller shippable pieces, chained so they land in order. and any bug it hits along the way becomes its own ticket.
❯how does it choose a machine, and can several run at once?
a dispatcher routes each ticket to any free, healthy machine that serves that repo. nothing gets stuck behind one offline box. connect as many machines as you like and tickets run in parallel. dependent tickets wait for their blocker to land, then go in order.
❯can my whole team use it, or is it single-player?
it’s built for teams. you share one organisation. everyone connects their own machines, work routes across all of them, and owner / admin roles control device and usage settings. you can also share a repo chat read-only across the org.
voluntold is built by voluntold. its own tickets, worked by its own agents.