labb

labb

A whole workshop of machinery for “make me an expense report app”

Your own AI assistant, self-hosted: a web chat for everyone at home, an agent in every terminal, and apps it builds for you, all on your own models.

Get labb
See it build an app
Make me an expense report app the leverwrite_app_filebuild_appopens it to checkyour app

labb: Expenses is running on port 9100. I opened it to check: receipts add up by category, and each month exports as a CSV report.

The web chat

Everyone gets a polished chat. You keep the server.

Run labb on one machine at home or at work, and everyone on the network can use it. Each person gets an account and their own chats, documents and memories. Replies can draw on your own files and the web in the same answer.

The labb web chat answering a question about yellow tomato leaves from the user's greenhouse notes and this week's forecast.
The terminal

And an agent in every terminal

labb reads and searches your code and runs commands with your approval, in any folder. It has a plan mode, goals it keeps working on until they're done, and /undo restores files when a turn goes wrong.

It shares the server's models, documents and memories, and its sessions show up in the web chat. Run labb connect and you can start work on that PC from the browser.

More on the terminal agent

labb in a terminal: asked why Docker rebuilds every dependency, it lists the project, reads the Dockerfile and pyproject.toml, and asks before an edit that installs the dependencies before copying the code.
An Expenses app running inside labb, with this month's totals by category, and the chat beside it where labb built it.
Apps and artifacts

Ask for an app, get one running

Say what you need and labb writes an app, builds it in a container of its own and opens it to check it works. Every build is a version you can go back to, and an app can have its own database and users.

In any chat, ask for a dashboard, a chart or a diagram, and you get an artifact with a link.

More on apps and artifacts

Put it to work

It keeps going after you close the tab

The agent editor: a Chef agent's instructions, model and tools, with a chat to try it.

Agents for each job

Give an assistant its own instructions, model, tools and documents. Try it as you edit, then share it with everyone on the server. More on agents.

The Tasks page: a frost check three times a week with its runs, a weekly reading list and a weekday news digest.

Tasks on a schedule

“Every weekday at 8, summarize the top news.” An agent does it, and the result waits for you in a chat, marked unread. See how scheduling works.

Studio: the Frost watch workflow on a canvas, with its finished run beside it.

Workflows in Studio

Describe a job in words or build it on a canvas, with branches, loops and approvals. Follow one from start to finish.

Your models, your network

One lab, wired together on your own hardware

labb starts on small local models through Ollama. When you want more power, an admin can turn on a cloud provider from the browser and choose who gets which models.

YOUR NETWORK Browsers, iPhoneweb chat, iOS app Each PCterminal agent labb serverone agent, one history Postgres + pgvectorchats, docs, memories Web crawlersearch, read pages Browserclicks for the agent Python sandboxno internet Appsa container each Artifactsbuilt offline Ollamachat + embeddings The web Cloud modelsoff until you say API key
With Ollama, your chats, documents and memories never leave your machines. Web searches and the agent's browser go out to the web, and neither can reach your network. Apps run in locked-down containers, and cloud models only come into it once an admin turns them on.
And there's more

Talk to it, teach it, wire it up

Voice mode

Talk hands-free, with speech models that run on your own machine.

Documents and memories

It answers from your files, links to the passages, and remembers what matters to you.

Integrations

Google, Microsoft 365, GitHub and Notion, any tool through MCP, and skills that teach it how you work.

The browser

It clicks through sites and fills in forms, and leaves passwords and payments to you.

Engines

Fork the agent's core, change its prompts and steps, and compare the results.

Your PCs

Start work on a PC at home from the browser, in the folders you allowed.

Build your own lab

You need Docker and Ollama. The server prints a setup link for the admin account, so nobody else can take it.

Get labb