All your coding agents.
One personal AI.

Jaz runs the coding agents you already use, on the plans you already pay for, and lets them work together, with memory, Tasks, a CRM and your accounts around them.

Claude CodeCodexGrokKimi CodeAntigravityMuse CodeOpenCode
lumen

Ask anything, or hand your assistant a task…

Agent for this thread

Make it yours

Every agent, as itself.

Jaz runs coding agents over the Agent Client Protocol, so you get the agent its lab built, with the subscription you already have. Seven work out of the box, and any other ACP agent plugs in.

Per thread Pick the agent for each thread and switch when a better one ships.

Think it through on a canvas.

Start a thread as a canvas and the agent lays its work out on a board beside the chat: notes, charts, tables, and files as live viewers, from 3D models and drawings to PDFs and web pages. Select anything to ask about it.

The board is a JSON Canvas file in the thread's folder, so every agent edits it with its own tools.

Canvas. A live board of charts, tables and notes beside the chat.

Agents that work together.

A bot is an agent with a name, a face and its own home folder, running whichever agent and model you choose. @mention who should answer, and let bots build on each other's work.

Long jobs go to background workers, and each bot shows up separately in Usage.

Hand off work A bot delegates a fix to Codex, reviews it, merges it and reports back.

Tasks and a CRM, built in.

Every agent knows your issues, people and deals. Agents file and close issues, build contacts and deals from your email and meetings, and draft the replies you owe.

tasks.jaz.chatcrm.jaz.chat

Tasks Issues, projects and cycles that agents plan and update.

Memory that survives the thread.

Jaz remembers what matters: who's who, what you're working on, how you like things done. Agents pick it up as they work, and Jaz dreams overnight, while you sleep. Every fact keeps a source you can check.

It's all plain files on your machine. Open them, search them, back them up.

Your accounts, in reach.

Connect Gmail, Calendar, Drive, Slack, Granola, WhatsApp and Telegram once, and every agent and bot can search and act with them. Any MCP server adds its tools for every agent, and skills are one $ away.

Sessions and tokens stay on your machine. Several accounts per service.

Connections. Accounts that agents search and act with.

Everything else they need.

Browsers, your desktop, live boards, Git, and a server that stays on.

MCP and skillsAny MCP server, MCP Apps in the sidebar, ⌘K search, skills one $ away.BrowserA side browser, your signed-in Chrome, or a background Chromium.
Computer useAgents open, read and click your desktop apps, once you switch it on.
VoiceCall a thread or a bot from a floating pill and follow the live transcript.
BoardsLive artifacts that refresh on a schedule, on any display.
GitCommit, push, open and merge pull requests from the thread.
Plan and GoalApprove the plan first, or set a goal it works toward until done.
Server modeRun the backend on a server and pair your devices to it.Colour themesUp to three linked hues, light and dark, and film grain.

A client for you. A server for the work.

Jaz splits in two. One half is the app you open — on any of your computers — to talk, review, and steer. The other keeps running on your Mac, a home server, or a small VM, taking in integrations and running routines with no window open.

Serveralways onIntegrationsLoopsMemoryYou

On your laptop — review, steer, and talk to agents.

Own your personal AI.

You should own your personal AI: the system, the memory, the data. Models come and go — Claude today, whatever is best tomorrow — but everything they learn about you stays in plain files on machines you own. This is too profound a technology to be locked inside someone else's tool. So Jaz is open source — yours to read, audit, run, and change.

Your machines. Your files. Any model. No lock-in.

If Jaz resonates, star it on GitHub.