AI that follows your rules, not its own

Transform the way you build LLM-based applications. AWMT provides a robust framework for creating AI agents that navigate through predefined logic, ensuring reliability and alignment with your business processes.

No more fine tuning. No more prompt engineering and hoping that things go as you wish.

1. Define your own types and logic rules

Easily map out your data structures, objects, and business processes using our interface, TypeScript SDK, or API.

// logic.ts
const awmt = new AWMT(process.env.AWMT_SERVER, process.env.AWMT_TOKEN);

const customer = await awmt.assert_object("customer", (field) => ({
	stripe_id: field.string(),
	name: field.string(),
	email: field.string(),
	phone: field.string(),
	company: field.reference("company"),
	payment_method: field.reference("payment_method"),
}));

const invoice = await awmt.assert_object("invoice", (field) => ({
	stripe_id: field.string(),

	customer: field.reference("customer"),
	amount: field.number(),

	good_for_us: field.boolean(),

	due_date: field.date(),
	paid_date: field.date(),

	pdf_link: field.string(),

	credit_note: field.reference("credit_note"),
	line_items: field.many_references("line_item"),
}));

2. Integrate AWMT into your app in minutes

Here is the simplest way to integrate AWMT with your Next.js app. Just tell it which objects it can user, and start interacting with your AI agent.

// actions.ts
import { AWMT_AI } from "awmt";

// Define the objects the agent is allowed to interact with.
	
export const AI = AWMT_AI({
	customer: true,
	invoice: true,
});

// app.tsx
import { useAwmtAI } from "awmt";
import { AI } from "@/app/action.ts";

export default function MyApp() {					
	const { 
		connected, 
		post_message, 
		can_post_message, 
		UI 
	} = useAwmtAI<typeof AI>();

	const { Messages, Goals, Forms, Choices } = UI;
	
	return <>
		/* Displays the messages exchanged between the agent and the user. */
		<Messages />
	
		/* Displays the goals set by the agent and their progress. Let the
		   user interact with them. */
		<Goals />
	
		/* Displays the forms generated by the agent to collect information
		   from the user. */
		<Forms />
	
		/* Displays the choices that the agent offers to the user to solve
		   ambiguous situations in one click. */
		<Choices />
	
		<Input 
			value={input} 
			onChange={v=> set_input(v.target.value)} 
		/>

		<Button /* ... */ />
	</>
}

3. Input complex queries or tasks

AWMT offers you three intuitive ways to communicate, ensuring you get exactly what you need, how you need it:

1. API Integration: For when you need full precision, our API allows you to submit structured queries and goal definition. Have complete control over what the agent does, and how it does it, ensuring your goals are met with accuracy.


2. Natural Language: Speak to AWMT like you would to a simple LLM. Whether you have queries or complex tasks to run, the natural language interface understands and executes your requests without the need for complex programming, while sticking to your logic.


3. Dynamic UI Components: Stay engaged with your tasks as they unfold. Our system generates interactive UI elements, allowing you to steer the agent in real-time, ensuring the outcome aligns perfectly with what you want.

These options make AWMT adaptable to your workflow, providing flexibility and precision when you need it most.

4. Watch the AI plan and execute goals