Runtime for real-time agents
Ship real-time AI agents faster. We handle socket communication, agent routing, and persistent state. So you focus on building product, not managing infra.
$ npm i @axarai/axar
axar.platform()
Everything between your users and agents.
- Built-in persistent messaging
- Real-time memory updates
- Context-aware message classification
- Ultra-fast multi-agent coordination
- Developer-first SDK
Build agents
core()
A super-lightweight framework for building real-time AI agents your way.
@model('openai:gpt-4o-mini')
@systemPrompt('Help customers with billing issues')
class BillingAgent extends Agent<string, string> {
constructor(routes: Routes) {
// Subscribe to relevant conversation routes
subscribe([routes.billing, routes.payment])
}
run(input: string): string {
return super.run(input);
}
}
Connect users
proxy()
Out-of-the-box client/server comms with low-latency semantic routing powered by Rust.
import { connect } from '@axarai/proxy';
// Establish a persistent socket connection
const channel = connect('axar.ai/v1', auth, conversation);
// Send a test message
channel.send({ type: 'ping' })
// Set up an event listener to handle incoming messages
channel.on('message', (msg: string) => {
console.log('Received:', msg);
});
Manage state
memory()
Persistent context storage for seamless state management across interactions.
import { context } from '@axarai/memory';
// Retrieve all previous messages from a conversation
const messages = context.get(conversation);
// Get a summary of user context
const summary = context.summary(user);
// Query specific information about the user
const result = context.query(user, ['user name']);
