The assistant takes the step for them.
Frigade guides users through your product in real time. With tool calls, it can also take the action for them, in your product and with their own permissions, so the workflow finishes instead of stalling on a step they didn't want to do.
Sometimes the best help is doing it for them.
Frigade already walks users through any workflow in context, highlighting the right controls as they go. For most things, that's exactly the help someone wants. But some steps are tedious, technical, or easy to get wrong, and showing the way still leaves the real work to the user.
For those, a tool call goes one step further. The same trained assistant performs the action itself, in your product, with the user's own permissions. The user asks, and it's done.
Real actions, inside your product.
Each action is a function you expose through the SDK. Frigade decides when to call it from the conversation. Valley, for one, uses Frigade to trigger team invites straight from a question instead of sending the user off to settings.
Invite a teammate
A user asks to add their team. The assistant sends the workspace invitation instead of walking them to the settings page.
Check a status
Look up an order, a request, or an account detail and answer with the live value, not a link to where it lives.
Submit a form or request
Kick off a request or fill a form on the user's behalf when they describe what they need in plain language.
Open a ticket with context
When a question genuinely needs a human, create a ticket in your support tool with the full conversation attached.
Retrieve a key or setting
Surface the thing the user is hunting for, like an API key or a configuration value, right in the conversation.
Handle higher-stakes actions
Put approvals in front of things like refunds, so the assistant can help without overstepping.
You decide what it's allowed to touch.
Nothing by default
No actions are enabled until you define them. The assistant can't do anything you haven't wired up.
Confirmation where it counts
Start with low-risk reads. Put a confirmation step in front of anything that changes data.
The user's own permissions
Tool calls run in the authenticated user's context, so the assistant can never do something the user couldn't do themselves.
Defined in your SDK
Each tool is a function you control, with its own inputs, guardrails, and scope.
Tool calls build on the rest of the assistant.
Actions ride on the same trained assistant that answers questions and guides users. Pair them with proactive Suggestions to trigger an action at the right moment, or route to a human through your support stack when a request needs one.