Agents use your accounts without seeing the keys

No. Agents work with placeholders, and the Laze proxy adds the real key to each request on its way to the service. The key is never in the agent's files, memory or output, so there is nothing for it to leak.

When a script on your agents' computer needs a key, it gets it for that one command only, and the key is hidden from the command's output. A request that comes from outside your agents' computer always asks the owner first.

For example, the Sales Agent can send an email from Mara's Gmail without ever holding her Gmail token. Every use goes into the audit log, with who asked, which agent and what the answer was.

Laze asks you first. You get a card in your feed and a notification on your phone or Mac, with Allow, Allow for 2 days and Deny. You can add a note, which the agent reads before it tries another way.

Allow covers that one request. Allow for 2 days lets it use the connection without asking again for up to two days. A card you don't answer is denied after 24 hours.

Agents shared with the whole team never use personal accounts. If you want a channel's agents to use one of your connections, you lend it to that channel for a while.

Each connection gets one of four levels. Auto lets agents use it whenever the task needs it. Checked has a second model read each request and lets it through when it fits the task, or asks a person when it doesn't.

Ask makes every use wait for the owner's yes, and Human only keeps agents away from it, so a person runs that step. A team might put GitHub on Auto, Stripe on Checked and the production database on Human only.

Admins can cap the level of any connection, so a member can't set a sensitive one to Auto. Some actions ask a person whatever the level: emails and messages to people outside the company always start as drafts.

Only the owner can see the value of a personal key, and only admins can see a workspace key. Before the value shows, Laze asks for a fresh sign-in or a check on your device, such as Touch ID.

The value shows for 30 seconds, and every view goes into the audit log. Keys are stored encrypted in the European Union.

For the most sensitive keys, you can seal a key to your own devices. Laze then stores only a locked copy and each use asks you. If you lose all your devices, the key is gone with them.

Laze spots it on the sender's device before the message goes out. If the message leaves Laze, for example to a Slack channel or to an agent running on someone's laptop, Laze asks the sender to name the secret and save it to their vault first.

Once saved, the channel shows a named vault entry instead of the password, and agents use it through the vault like any other key. In a message to an agent that runs in Laze, the key is saved straight away and a small label takes its place.

Laze also checks for stray secrets every day and hides the ones it finds. For example, if Tom pastes an API key into his chat with the Ops Agent, he sees a label saying the key is in his vault, and the key never shows in the chat.

Gmail, Google Calendar, Google Drive and GitHub connect with one click. Featured services such as Notion, Linear, Sentry, Stripe and Attio connect as MCP servers, and so does any server in the MCP Registry.

For anything else, you save an API key or a password in the vault, with a name agents use to ask for it. Every connection, whatever its kind, follows the level you set for it.