Google Workspace
Gmail · Calendar · Drive · Contacts · Tasks · Meet · Chat
Integrations
A droid is only useful if it can reach the same places you can. So it lives on the channels your team already uses, connects to your accounts under your own permissions, and falls back to a browser for the software that never got an API.
Every one of these is two-way, not a notification feed. A droid picks up a thread on email, carries the same work on over a phone call, and remembers both.
One consent screen, scoped to the surfaces you pick. A droid reads and writes as you, so it inherits your access rather than needing an admin to widen anything.
Gmail · Calendar · Drive · Contacts · Tasks · Meet · Chat
Outlook mail · Calendar · OneDrive · Contacts · Teams · SharePoint · To Do
A droid searches this catalog mid-task and asks you to authorise what it needs, so nothing here has to be wired up in advance.
1,045 apps.
Each connection is scoped and revocable. Credentials are held encrypted and swapped in at call time, so a droid can use a connection without ever reading the token behind it, and the detail is on the security page.
Connections belong to the person who authorised them, which is what keeps a droid's permissions the same as yours.
Most platforms have exactly one answer to "do you support this tool?" and it's a list. We'd rather a droid exhausted every route before telling you no.
You authorise a droid the way you'd authorise a new hire, through the app's own OAuth screen, or with an API key you paste in once. Credentials are held encrypted and swapped in at call time; the model never sees the secret itself.
Nobody sits down and pre-wires a hundred connectors. A droid searches the catalog mid-task, and if what it needs isn't connected yet it asks you to authorise it rather than guessing or failing quietly.
Plenty of the software a team depends on has no usable API: an old portal, a supplier's booking system, a council website. A droid gets its own desktop and browser and clicks through it the way a person would.
Droids run real code. Given an endpoint and a token, one writes the request, installs whatever library it needs, and gets the data. Anything worth keeping is stored as a reusable function.
Work arrives as files far more often than as API calls. Anything a droid receives gets parsed into something it can query, rather than stuffed into a prompt and half-remembered.
CSV and Excel parsed into queryable tables, not pasted into a prompt
PDFs and office files read page by page, including scans
Pulled off inbound mail and ingested without anyone downloading them
Read from the folders you've granted, through your own connection
Most likely, and there are three routes. Common SaaS tools are in the connected-app catalog and take one OAuth click. Anything with a public API can be reached directly by a droid writing the call. Anything with neither still works through the browser. If it's a tool you'd like first-class support for, tell us and we'll look at building it properly.
We do, encrypted, and they are injected at the moment of the call rather than handed to the model. A droid can use a connection without ever being able to read the token behind it. There's more detail on that architecture on the security page.
Yes. Connections belong to the person who authorised them, so a droid acting for you uses your access and not a shared service account. That's what keeps its permissions the same as yours.
The droid notices, stops, and tells you which connection needs re-authorising rather than reporting the task as done. Silent failure is the thing we work hardest to avoid.
If it's reachable over HTTP, yes, a droid can call it directly and you can store the pattern as a function for reuse. If it's only reachable from inside your network, the browser route on a linked machine usually covers it. Talk to us about the awkward ones.
Point a droid at your own stack, or read a worked example. Each one lists the apps and channels it runs across.