Claude Code · OpenRouter · Open source

KALAPU

Multiple projects. In parallel.
From your phone.

Kalapu runs on your always-on machine and orchestrates autonomous AI agents — coding ones and persona-only ones — across every one of your projects 24/7. Native kanban, chat with each agent, on-demand code review, token + USD cost per ticket. Bring your own LLM (Claude Code or OpenRouter). Send work and review changes from a mobile-first PWA — your phone is the office.

Read the article

How it works

From idea to production.
Without sitting at your desk.

Install on your always-on machine

A single command sets up Kalapu on any Mac you leave powered on — laptop, Mac mini, or old desktop. This becomes your agent hub.

Connect your repos (or just personas)

Add each GitHub repo to kalapu.json, or define a persona-only agent (a copywriter, a designer) with no repo. Each one runs on the LLM you pick: Claude Code or OpenRouter (Kimi, MiniMax, DeepSeek, Gemini, Claude via OR).

Send work from anywhere

Create a ticket in the native kanban (or drop a GitHub issue), chat with an agent from the dashboard PWA on your phone, or SSH in. Any channel works — the work starts in minutes.

Review from your phone

Live logs, kanban progress, and pull requests land in the mobile dashboard. Approve, merge, or redirect from wherever you are.

The office

Every agent has a desk. You can see who's working.

A pixel-art scene that mirrors what your agents are actually doing. Working agents walk to their desks. Down ones slump. The metaphor clicks on first glance — no legend needed.

Oficina Kalapu 14:42:08
Working Idle Down
K KALAPU kalapu HEALTH healthcheck titofree-v2 fastline bgb-v2 × strombapa mycvify sparkseo nico KALAPU · 6 AGENTS · 4 WORKING

Each agent's name, position and activity is live. The healthcheck wall shows real metrics. The walking agent is one currently dispatching a task.

The model

Always on. Always reachable.

Your machine is the hub — it never sleeps. You don't have to be sitting in front of it to make progress. Talk to it from anywhere, any way you prefer.

You · from anywhere

Kanban ticket

Create it in the native kanban (or drop a GitHub issue if you prefer). Auto-dispatches the moment it hits Todo.

Chat from the PWA

Streaming chat with any agent — coding or persona-only — from the installable dashboard on your phone, with image paste and markdown.

Terminal & SSH

Jump into any tmux session when you want full control.

live

Your always-on Mac

agent: project-alpha
agent: project-beta
agent: project-gamma
issue-watcher
dashboard :3000

Work · back to you

Pull requests

Reviewable diffs with the plan, the changes, and tests.

Mobile dashboard

Real-time status, kanban, and logs — all from the browser.

Ship signals

Deploys go out on their own when you've set an agent to fast mode.

Agent modes

You set the level of control

FAST

Ship directly

The agent implements changes and pushes directly to production. Maximum speed, minimum friction.

PLAN

Propose first

The agent drafts an implementation plan and waits for your approval before touching any code.

DEFAULT

Branch & PR

Creates a branch, implements the feature, and opens a pull request. You review before merging.

HUMAN

Keep it manual

Add the human label and the watcher skips it. Reserve the task for yourself or a teammate.

Dashboard

Everything in one place, from your phone

Pixel-art office where each agent lives at a desk. Chat with each one. Track tickets in a local kanban. Trigger a code review with one tap. All from an installable PWA on your phone.

  • Office tab: pixel-art scene that mirrors who's working, who's idle, who's down
  • Native kanban with streaming chat per agent, image paste, markdown render
  • Roles built in: admin, product manager (create + close), viewer (read-only)
K
Kalapu running
njpa382 · admin
Oficina Chat Tasks Agents
Agente: Todos titofree-v2 fastline mycvify
Done 1

#7Fix login token refresh on iOS

👤 fastline
⎇ a2c4f1d PR #18 ⚡ 8.4k · $0.04
In Review 1

#1Use calendar for manual premium assignment

👤 titofree-v2 fast
⎇ 9498816 PR #21 💬 3 ⚡ 14.3k · $0.083
In Progress 1

#9Add deep link to share an offer

👤 titofree-v2
⚡ 5.1k · $0.024
Todo 2

#10Refactor onboarding video state machine

👤 bgb-v2 plan

#11Generate CV preview thumbnail server-side

👤 mycvify

Why Kalapu

Designed for how solo builders actually work

Bring your own LLM

Claude Code or OpenRouter — Kimi, MiniMax, DeepSeek, Gemini, Claude via OR. Pick per agent. No Anthropic plan required.

Runs on your machine

No cloud, no backend, no third party. Your code stays where it is. Kalapu is orchestration around tools you already trust.

Native kanban + chat per agent

Local kanban with auto-dispatch on Todo, re-dispatch when you comment on In Review, streaming chat with each agent. No more dependency on GitHub Projects.

Code review on demand

One button on every In Review ticket. Pick the reviewer model, get a structured review posted as a ticket comment and on the PR. Optional, cheap, catches what the coder model missed.

Tokens + USD per ticket

Every dispatch and every review adds to the ticket's cumulative cost. You always know what each piece of work burned, in tokens and dollars.

PWA, mobile-first

Installable on your phone's home screen. Bottom-sheet menu, drag-and-drop kanban on touch, image paste in chat. Your home office no longer needs to be home.

Who it's for

Built for people who build alone

Solo founders, indie hackers, freelancers, and small teams who want to move fast across multiple projects — without hiring a team or burning out.

projects in parallel

agents working

subscription needed

FAQ

The obvious questions

Do I need special hardware?
No. Any Mac that runs Node.js + your chosen LLM CLI (Claude Code or OpenCode) works — a laptop, a Mac mini, or an old desktop in a closet. The only real requirement is that it stays powered on and connected to the internet.
Do I really have to leave my machine on all the time?
Yes — that's the point. An always-on machine is what lets agents pick up work at any hour and lets you reach them from anywhere. Kalapu installs a LaunchAgent so everything comes back up automatically after a reboot or power cut.
Where does my code run? Does anything leave my machine?
Agents run locally, in tmux sessions on your machine. The only outbound traffic is what your LLM CLI (Claude Code or OpenCode/OpenRouter) and GitHub already do on their own: prompts to your chosen model's API, pushes to your own repos. There is no Kalapu cloud, no backend, no telemetry.
Which LLM do I need? Do I need a Claude plan?
You pick. The installer asks whether you want Claude Code (Anthropic plan) or OpenCode + OpenRouter (Kimi, MiniMax, DeepSeek, Gemini, Claude via OR — pay as you go). You can override per agent later, so a typing-heavy coder agent might run on Kimi while your code reviewer uses Claude Sonnet. The dashboard tracks tokens + USD per ticket either way.
Can I talk to the agents from my phone or from another computer?
Yes. The dashboard is an installable PWA — drop it on your home screen on iOS or Android and it works like a native app. Chat with any agent (coding or persona-only), check tickets in the kanban, trigger a code review, all from anywhere. Claude Code agents also still connect to the Claude app on phone or desktop if you have a plan.
What happens if an agent breaks something?
Per-agent modes let you decide how much trust each one gets. Default mode opens pull requests for human review. Plan mode pauses for approval before writing any code. Fast mode ships directly — use it only where you're comfortable with the blast radius.
What if I want to do a task myself instead of handing it to an agent?
Add the human label to the issue. The watcher ignores any issue with that label — even in Todo — so you or a teammate can pick it up without an agent touching it. Remove the label any time to hand it back.
How many projects can Kalapu handle?
As many as your machine and your Claude plan can sustain. Each project is just one extra entry in kalapu.json.

Ready to let agents do the work?

Kalapu is under active development. Get in touch to learn more or request early access.