Let users finish the task by asking.
Frigade learns actions as it explores your product and turns them into Skills. Review and enable the ones you want your assistant to run, without writing a function for each one.
Appointments
Put repeat tasks within reach.
Looking up an order or changing a setting can mean finding the right page, record, and control. With an enabled Skill, your users can ask the assistant to do that work from the conversation.
Frigade discovers supported actions while learning your product. You review what it found and choose what to make available, so you can add actions without defining each one in code.
See the assistant in action.
Watch recorded demos of Frigade in Jira, Spotify, and Hacker News.
From discovery to a finished task.
Frigade discovers actions
As Frigade explores your product, it identifies actions it can turn into Skills and adds them to your dashboard.
You review and enable them
Check what each Skill does and the inputs it accepts. Discovered Skills start off, and you choose which ones become available.
Your users ask for the result
The assistant uses enabled Skills to look up information or make a requested change, then responds with the result.
Help with the work already in your product.
The Skills available depend on what Frigade discovers in your product. These examples show how an enabled action can help in different workflows.
Skills retrieve data and run actions. Generative UI presents the answer as tables, record links, or choices in the conversation.
How Generative UI worksChoose what the assistant can do.
Self-host with your own LLM keys
Run Frigade in your environment when your data requirements call for it. SOC 2 Type II and GDPR support your security review.
Review before enabling
Discovered Skills start off. Inspect each one in the dashboard, enable it when you're ready, and turn it off when you need to.
Use the signed-in session
Product Skills run in your user's browser using their current session. Keep permission checks in your product's API.
Ask before making changes
The assistant is instructed to ask for approval before running a Skill that creates, updates, or deletes data.
Add the actions you want to define yourself.
For an action Frigade hasn't discovered, or one with your own business rules, define a tool through the SDK. You write the function and describe when the assistant should use it.
Start with the Skills Frigade discovers, and add a custom tool where you need more control.
Your function does the work
Connect a tool to your API and return the result you want the assistant to use.
Available alongside Skills
Your tools join the assistant's enabled Skills, so it can choose an action from the conversation.
Your code enforces the rules
Validate inputs and enforce permissions in your API. Add an approval check to actions that need one.
assistant.ts (example)
TypeScriptFrigadeAssistant({
// ...your existing config
tools: {
getOrderStatus: {
name: 'Get order status',
description: 'Look up an order by its ID.',
inputSchema: {
type: 'object',
properties: {
orderId: { type: 'string' },
},
required: ['orderId'],
},
execute: async ({ orderId }) => {
if (typeof orderId !== 'string') {
throw new Error('An order ID is required.');
}
// Replace with your API, including its access checks.
return api.orders.get(orderId)
},
},
},
})Give every action a useful answer.
Skills do the work. Generative UI presents the result, using what Frigade has learned about your product.