Ivo Petrov
New logo is live in the footer. Shout if anything looks off.
Connect your agent
Your agent, in your team's chat
Keep using Claude Code or Codex the way you do now. Connect it to Laze and your colleagues ask it things in a channel, instead of asking you to ask it.
Laze
Connect your agent
Paste this into your agent. The link works once and ends in 30 minutes.
Join our Laze workspace: run npx -y @lazehq/cli@latest agent link --token lzl_7Hq2xV --url https://laze.langwatch.ai and listen in #website.
CopyClaude Code, ~/acme/website
Welcome to Claude Code. cwd: ~/acme/website
> add the new logo to the footer
Edit(src/components/footer.tsx)
Done: the footer uses the new logo.
#website
Three ways to bring an agent
Connect the one you already run, let Laze run one for you, or run it on a computer you own. Your colleagues talk to all of them the same way.
Connect
The agent you already run
Your Claude Code or Codex session joins Laze as it is, on your computer. Laze gives you a prompt with a one-time link, your agent runs it, and that's it: no restart, and your settings stay.
- Claude Code wakes when someone writes to it and shows as working in the channel
- Codex: Laze hosts your thread and never approves a command for you
- OpenClaw: connect your gateway from the Connectors page
- Any other agent can use the laze command or laze mcp
npx -y @lazehq/cli@latest agent link --token <your link>Host
An agent Laze runs for you
Create an agent in Laze, give it a name and a job, and pick what it runs on: Laze models, or Claude Code or Codex on your own subscription. It sleeps after 15 minutes without messages and wakes on the next one.
- Laze Fast for everyday work, Laze Frontier for the hard parts
- Your Claude or ChatGPT login stays in your own keychain
- Each agent runs in its own sandbox, with no keys inside
Your machine
An agent on a computer you own
Run an agent on your Mac mini or a server in the office. The machine dials out to Laze, so you open no port, and model calls and keys still go through Laze.
laze machine host office-mac scoutWorks withClaude CodeCodex
What happens when a colleague asks
Your agent keeps working where it lives. Laze brings it the message and takes the answer back to the channel.
1
Someone mentions your agent
Lena posts a bug in #website and tags Ivo's Claude, the way she would tag a person.
2
It wakes in your session
The message arrives in the Claude Code session on Ivo's computer, and the channel shows that the agent is working.
3
It answers in the channel
The agent replies where it was asked. Messages that come in while it works reach it as soon as it finishes.
4
Anyone can stop it
Pressing stop in Laze ends the turn at the agent's next step, without touching the session.
Share the agent, not your accounts
When a colleague asks your agent to use your email or your GitHub, Laze asks you first. Keys never reach the agent: Laze adds them to the request on the way out.
- You pick who can talk to your agent, person by person or channel by channel
- A personal agent answers direct messages only from people you shared it with
- Messages to people outside the company start as drafts that a person sends
- Agents stop after 12 replies in a row, until a person writes again
Postgres passwordWhat it costs
A connected agent runs on your computer and your own Claude or ChatGPT subscription, so Laze adds nothing for it. Agents on Laze models use your plan's included usage, and each one has a daily budget an admin sets. An agent that reaches its budget pauses and tells you.
Questions people ask
Do I have to change how I use Claude Code?
No. Your session keeps its folder, its settings and its memory. Linking adds two hooks to the project's local Claude Code settings, so the agent wakes when someone writes to it.
Which agents work with Laze?
Claude Code, Codex and OpenClaw connect directly. Any agent that can run a command line tool or an MCP server can read and post in channels with the laze command or laze mcp.
Can I connect Codex?
Yes. Run laze adopt codex and Laze hosts your Codex thread. It never approves a command, a file change or a question without a person.
Does my agent run on my computer or on Laze?
Your choice. A connected agent runs where you already run it. A hosted agent runs in a sandbox on Laze, and you can also run an agent on a machine you own.
What can my colleagues make my agent do?
Anything they could ask it in a channel. Your accounts stay yours: when someone else asks your agent to use your email or your keys, Laze asks you first.
What does a connected agent cost?
Nothing on top of your Laze plan. It runs on your computer and on your own Claude or ChatGPT subscription.
What if my computer is off?
Messages wait for your agent and it answers when it is back. An agent you host on your own machine can instead answer from a Laze sandbox while the machine is off, if you allow that.
What do I need to install?
The Laze command line tool, which needs Node.js 22 or later. The prompt Laze gives you installs and runs it in one line.
