The parts, and what teams build from them.

Everything below is a building block. Your team snaps them together into apps, agents and routines, in its own words, and no two companies end up with the same Kico.

Apps

Any app your team can describe.

One line in the chat starts a project. A dev agent builds it in its own sandbox, runs it next to the chat and checks it before it says done.

Any stack
The agent is root in a Debian sandbox: Vite, Next.js, Django, FastAPI, PHP, or a plain script.
Checked in a real browser
Install, start, and a headless browser that opens the page, reads the console and takes a screenshot.
Several apps in one project
A web app, its API and a worker, each on its own port, each with its own preview and checks.
Bring a repository
Import from GitHub or any Git server; the agent works out how to run it.
Preview on a phone
Desktop and phone preview, logs for apps without a screen, a zip of the whole project.
Share like a document
Can use or Can edit, for members, the organization, or a link with email codes.
Push to GitHub
Every push goes through Kico with the credentials of whoever is working. No token in the sandbox.
Stop or redirect any time
Messages you send while it works are picked up after the current step.
Assistant

Kico knows your work.

Your assistant lives in a panel on every page. It knows your projects, organizations and colleagues, and works while you're away.

Memory you can read
A short profile every project reads, facts it found, a summary of every task. Fix or delete any of it.
Schedules and events
"Every weekday at 8" or "when checks fail". Each run is a task you can open.
Talks to colleagues
Messages to colleagues land in their bell, from your assistant, and by mail when they're away.
Runs your organization
Invites, roles and settings from a chat, through the same permission checks as the screens.
Knows where you are
Which page you're on, which project, who is online.
One kill switch
Stops everything it started, at once.
Agents

A helper for every job.

Describe the job in your own words. A model asks what it needs to know and drafts the agent while you watch; you test it before it's real.

One job, its own tools
Web, projects, data, mail, memory, files, connectors. Nothing it wasn't given.
Starts on its own
A schedule, an event, a new mail, or when you ask.
Checks with you
For its first runs, or always, what it would send waits for you: send, change or decline.
Its own inbox
Every agent has an address. Mail it, copy it, give it to a supplier. Try ours: demo@agents.kico.dev.
Sends without asking, to whom you say
An address, a domain or your colleagues. Everyone else waits for you.
Templates
Share a setup with your organization. Never its conversations or data.
Browser

A browser that does the clicking.

Kico has its own Chromium with your sign-ins. Watch it live, and take it over when a step needs you.

Hands over for sign-ins
A captcha or a login: it hands you the browser and waits for Done.
Out through your home
Sites that distrust datacenters see your home connection, over Tailscale.
Sign-ins that stay
Cookies live in the assistant's private storage, from one task to the next.
From a laptop or a phone
Mouse, keys, touch and paste, file uploads, dialogs and new tabs.
Data

Your databases, read-only, and understood.

Connect PostgreSQL, MySQL and MariaDB, SQL Server, Oracle or a SQLite file. A model describes every table, so agents find the right one by meaning.

Read-only by construction
Each engine its own way: read-only transactions, one statement per call, row limits and timeouts.
Credentials stay put
Apps and agents query through Kico. Passwords never reach a sandbox.
SSH jump hosts
For databases that aren't on the internet.
Every query logged
Who asked, what ran, how many rows.
Models

Any model, down to your own GPU.

Pick a model per project, per agent and for background work, with your key, your plan or your organization's. Or run your own.

Every major provider
Anthropic, OpenAI, Google Gemini, OpenRouter, xAI, NVIDIA, Reflection AI.
Your own server
vLLM, LM Studio, ollama or llama.cpp, at home through Tailscale without opening a port.
Dictation that stays home
Model servers can transcribe your voice too.
Learns your work
When a stronger model steps in, Kico keeps the case; your GPU trains on the week's cases over the weekend.
Every token counted
Per day, organization, person, project and model.
Teams

Built for a whole company.

Organizations, roles and invites, with Kico greeting every new colleague and an audit log behind every change.

Roles
Owner, admin, member, viewer, in each organization you're in.
A welcome from Kico
New colleagues land with Kico already talking: who invited them, what they can do.
The bell
Messages, results and requests that need you, also by mail when you're away.
Audit and usage
Changes made by admins and by Kico on your behalf are logged, and every token is counted.

And whatever you think of next.

Most of what teams do with Kico wasn't on any list. Tell it what you need.

Kico is invite-only. What you write here becomes your request.

Or book a demo and we'll build something of yours live.