// long-horizon work, on hardware you own
hev factory orchestrates a crew of coding agents on a Mac you own. You approve a plan, and it opens the pull requests.
The whole machine is Apache-2.0 and free. Pro is that machine provisioned, on hardware that never sleeps.
$ brew install hev/tap/factory
$ factory
factory: cloning into ~/workspace/factory
reception-up: receptionist on duty in 'reception'
↓ the picker opens. reception is the top row.acme ↵ attach ^d details ^g tell gaffer ^x stop one · type to filter
💁 reception the front desk — ask anything
── sub-agents ──
● gaffer-acme acme claude working dispatching the index worker
●! worker-acme-index acme ~index claude working npm test has failed the same way …
○? worker-acme-search acme HEV-14 rfc-search claude waiting asks which index to rebuild first
🚨 stop the line 3 agent(s) in 3 sub-agent(s)
Everything the factory is doing is a row. A dot says whether an agent is
streaming output or sitting idle, a ! marks the worker in
trouble and a ? the one waiting on an answer from you, and
Enter drops you into any of them mid-task. The last row is the andon
cord.
# the whole system
Everything in the repository is one of them made concrete: a Go
binary for the picker, a markdown charter for reception, a markdown
file in plans/active/ per RFC, and a markdown contract
plus a launchd plist per gaffer. All durable state is a file under
~/.factory.
The front door, and most days the only screen you need. It lists every session the factory owns, with the plan step each one is on and what it is doing right now. Enter attaches to a row, typing filters the list, and the last row stops every agent in the factory.
Read the docs
The desk you talk to. On the first run it walks you through your first factory, then answers for the whole floor. Bring it a half-formed idea and it argues back until there is an RFC worth approving. It never dispatches, never merges, and never approves.
Read the docs
The unit of approval. Every change starts as a markdown RFC you wrote with reception or by hand, and you approve it by merging the pull request that adds it. Connect Linear and the RFC is an issue instead, approved by moving it into one state, which is a tap from a phone. Either way, nothing gets dispatched that does not trace back to an RFC you approved.
Read the docs
The parent agent for one factory. Each beat reads the plans you approved, breaks them into tasks, dispatches a worker per task into its own tmux session, and reports what needs you. Its instructions are a markdown file you edit.
Read the docs# multi-repo · one harness
A factory's scope is a list of repositories. One approved RFC becomes issues and pull requests across every repo it touches, and each repo gets its own concurrency lane so two agents never work the same tree at once.
This is the case a single agent session handles worst. A feature that spans an API, a client, and a docs site is three working trees, three test suites, and an order the changes have to land in.
Every agent runs claude: the desk, the gaffer, and
every worker it dispatches. That is one thing to install and one
subscription to hold. What the gaffer chooses per task is the model
and the reasoning effort, reaching for the heaviest model where the
shape of the work is unclear and a smaller one where the answer is
already known.
It ships no model and never sees your API keys. Every agent is a tmux session, so you can attach to any of them mid-task and take over by typing.
# how a change happens
Every worker runs interactive in its own named tmux session, so you can attach to one mid-task and take over.
# hev factory pro
The open version needs a Mac that never sleeps, and it runs every
agent as you, on the one GitHub account gh auth already
is. Pro is that machine, provisioned: a mini that stays awake, an
account per role so the gaffer is not you, and your own subscriptions
signed in on it. Leave an email and we will take it from there.
