01
The harness, one key away
Every session is a real Claude Code, Codex or OpenCode session. Attach to it from the workbench, see exactly what it shows and type into it. cland adds structure on top; it never walls the harness off.
Learn more →OPEN-SOURCE CREW RUNNER · APACHE-2.0 · macOS AND LINUX
All your subscriptions.
One team.
One coding crew across every AI plan you pay for. cland runs Claude Code, Codex and OpenCode together: a strong model leads, cheaper and open models do the legwork, and you make the decisions.
$
brew install kealman/tap/clandHomebrew, the install script and source: all install paths.
◐coordinatorClaude Codeopus◐implementer-1OpenCodeqwen3-coder◐implementer-2Claude Codesonnet◉reviewerCodexgpt●qaCodexdemo passed
▲ coordinator asks you
Epic "Checkout flow" is planned in 3 stages, 7 tasks. Start it?
1 approve · 2 change the plan · 3 not now
You pay for three AIs.
They never work together.
Limits hit one at a time
Claude runs out in the middle of a task while your ChatGPT plan sits idle.
Every agent works alone
One terminal per harness, and you are the one copying context between them.
Frontier models do grunt work
Your best model writes boilerplate and burns the plan you need for the hard parts.
A crew you design
Choose the roles, how many of each, the harness and model every one runs on, and how work moves between them: lanes, review rules, QA policy, who approves the plan. One hub connects the sessions; one tracker lives in your repo as Markdown. A few crews people run:
Claude, ChatGPT and open models, each where it is strongest.
Coordinator×1
Claude Code · opus
Plans the release, splits epics into tasks, launches the crew and brings you only the decisions.
Implementers×3
OpenCode · open models
Each takes one task in its own worktree and works it to a reviewed branch.
Reviewer×1
Codex · gpt
Reads the change with another vendor's eyes and records a verdict.
QA×1
Codex · gpt
Runs the brief against the real thing and accepts the epic or sends it back.
You
with your Aide
Approve plans when you opt in, answer the asks, accept the release. Your Aide speaks for you when you are away.
+ your own role
A docs writer, a security reviewer, an epic lead. A role is a few lines in .cland/roles.toml.
A single Claude plan: the strong model leads, the faster one does the work.
Coordinator×1
Claude Code · opus
Plans and leads; the only session that spends the top model.
Implementers×2
Claude Code · sonnet
Work tasks in parallel, each in its own worktree.
Reviewer×1
Claude Code · sonnet
Reviews every branch before it can merge.
You
with your Aide
Approve plans when you opt in, answer the asks, accept the release. Your Aide speaks for you when you are away.
+ your own role
A docs writer, a security reviewer, an epic lead. A role is a few lines in .cland/roles.toml.
Frontier model for judgement only; the legwork on free and open models.
Coordinator×1
Claude Code · sonnet
Plans and decides what goes where.
Implementers×4
OpenCode · free models
Many cheap hands on small, well-cut tasks.
Reviewer×1
Codex · gpt
One strong reviewer keeps the bar.
QA×1
OpenCode · open model
Runs the brief and the checks.
You
with your Aide
Approve plans when you opt in, answer the asks, accept the release. Your Aide speaks for you when you are away.
+ your own role
A docs writer, a security reviewer, an epic lead. A role is a few lines in .cland/roles.toml.
One Coordinator to lead, one Implementer to work through tasks.
Coordinator×1
Codex · gpt
Plans the work and brings you the decisions.
Implementer×1
Codex · gpt
Works one task at a time and reports back to the Coordinator.
You
with your Aide
Approve plans when you opt in, answer the asks, accept the release. Your Aide speaks for you when you are away.
+ your own role
A docs writer, a security reviewer, an epic lead. A role is a few lines in .cland/roles.toml.
Get more from every plan you pay for
See the usage windows Claude Code and Codex report in one place; OpenCode has no limit data. Pick the account each role runs on. When a plan runs low, the work moves to the next one: you choose where, in one answer.
- Frontier models only where judgement matters
- Open and free models for the legwork
- Your own logins, the official CLIs, no accounts rotated behind your back
In the terminal, the browser or your pocket
Start the crew from Neovim, follow it in the browser, answer an ask from your phone. Every client talks to the same hub, so nothing is lost when you switch.
Built for shipping, not demos
02
Every plan's limits in view
The usage windows Claude Code and Codex report, account by account. OpenCode has no limit meters. A limited account stops new launches, and you choose where the work goes next.
Learn more →03
Your crew, in Telegram and Discord
Follow and steer the crew from a private Telegram chat or a channel in your own Discord server. General goes both ways; asks and Codex and OpenCode approvals arrive as cards. Send photos, files and voice notes. Pair through cland setup on the machine running the hub.
Learn more →04
Only what needs you
Questions travel up the chain of leads, and only the Coordinator asks you. The inbox holds the asks, approvals, plans and requests waiting on you, each answered with one key, from the terminal or your phone.
Learn more →05
A gate before every merge
A task lands on dev only with clear verdicts from other sessions, current for the branch's head, and every acceptance criterion done and verified.
Learn more →06
Your tracker, in your repo
Releases, epics, tasks, specs and questions are Markdown files committed with your code. Configure the container ladder, collect loose work in Bugs, Tech debt, Follow-ups and Ideas, and import your GitHub issues in one command.
Learn more →07
Topics, not one noisy chat
Talk to the Coordinator in topics bound to the work, each with a status that says who is waited on. General is there for everything else, and agent-to-agent chatter folds away into its own read-only topic.
Learn more →08
Plans with stages and lanes
Every epic gets ordered stages, parallel lanes and a QA policy before work starts. Approve each plan yourself, or let the Coordinator start and tell you.
Learn more →09
Verification for any stack
Your own prepare, check and test commands, through one queue that never overloads your machine. Or run them on fresh cloud machines in parallel, under a daily spend cap.
Learn more →10
Artifacts in the thread
Reports, diagrams and pages a session shows you arrive in their topic, and HTML renders inline in the browser. Keep the ones that matter, or promote them into the repository.
Learn more →11
Releases from the item
From an idea to GitHub: grill it or chart it with a wayfinder map, write the spec, plan the epics, build the tasks in parallel lanes through the merge gate, demo each epic, and publish with one atomic push once you approve.
Learn more →Get started in two minutes
$ brew install kealman/tap/cland
$ cd my-project
$ cland init && cland setup
$ curl -fsSL https://cland.app/install.sh | sh
$ cd my-project
$ cland init && cland setup
$ git clone https://github.com/Kealman/cland-v0 cland
$ cd cland && npm ci && npm link
$ cd my-project
$ cland init && cland setup
Setup finds the harnesses you have, offers the ones you don't, signs you in with their own login commands and opens the workbench with your Aide, who takes it from there.
macOS or Linux · at least one of Claude Code, Codex, OpenCode · Read the quickstart →
Questions
Is this allowed by my subscription's terms?
cland runs the official claude, codex and opencode CLIs under your own logins, exactly as you would by hand. It never reads your tokens to call a provider's API, and it never rotates accounts.
Which subscriptions work?
Whatever the harnesses sign in with: Claude plans for Claude Code, ChatGPT plans for Codex, and every provider OpenCode supports, free and open models included.
Do I need all three harnesses?
No. With one, the whole crew runs on it. Each one you add widens the team and spreads the load.
Does my code leave my machine?
To the model providers you choose through their CLIs, and to your DigitalOcean account if you enable cloud verification. cland has no telemetry. Its free-model warnings are a prompt to check your provider's data policy.
Neovim or the browser?
Both, on the same hub. The terminal workbench needs no Neovim experience: cland brings its own, with every key in a palette.
What does it cost?
cland is free and open source under Apache-2.0. You pay for your AI plans and any cloud verification machines you enable.
From the build log
cland is built by a cland crew. After each release we write down what it took, what it cost, what broke and what we changed.
v0.3.1: a fifth of the memory, and no silent Coordinator
v0.3.1 cuts the cland hub to about a fifth of its memory on our own project and makes sure a message to the Coordinator never silently goes undelivered. 25.5 hours from a voice note to the release, steered from a phone.
v0.3: the crew in your Telegram and Discord
v0.3.0 lets you follow and steer an AI coding crew from Telegram or Discord: General both ways, asks and approvals as cards, voice notes as answers. Seven epics, every demo clear on its first run, and a live run on real bots that found the bugs the fakes could not.
v0.2 and v0.2.1: the crew moved into the browser, and a free model landed six tasks in a row
v0.2 put the whole workbench in the browser and on the phone: 12 epics, every demo clear on its first verdict. v0.2.1 then ran a free model in every crew role on real tasks. It landed its work with no model-caused rescue. What failed was our ability to prove it.