The backend for end-user agents

An agent for every one of your users

Each end user gets their own isolated, separately metered agent with its own workspace, sandbox, memory and quota. Start with an assistant that chats; switch on capabilities in configuration and it grows into one that gets things done.

One way up

From chat to an agent that does work

Same channel, same component, same quota. Upgrading is a configuration change, not a rewrite.

01
Chat
An assistant inside your product that answers each user in their own conversation.
02
Knowledge
Answers grounded in your documents through a project knowledge base.
03
Work
Files, data, code execution and your own systems, so the agent carries tasks through for the user.

What you get

Five things every user's agent comes with

Together they let the agent act as that user and carry a task through, not just answer.

Your systems

Tools call your APIs and third-party services with that user's credentials, within that user's permissions.

A place to work

Its own files, database and code sandbox, so it hands back results, not just replies.

Memory

Remembers the user and the conversation across visits, with streaming and reconnects handled.

Quota control

Credits for each user on your existing login, charged up front. No overruns.

Isolation

One user's history, files, sandbox and memory are never visible to another.

Integration

One component, one key

Drop the React component into your app. Identity, sessions, data, quota and outbound calls are wired up behind it.

Trusts the login you already have
Every resource scoped to one end user
Change instructions, tools or model without a redeploy

import { InstinctAgent } from '@instinct-build/react'

export function Assistant() {
  return (
    <InstinctAgent
      publishableKey="ist_pk_…"
      agent="default"
    />
  )
}

Try the component in the playground

Build the product, not the agent backend

Configure agents, channels and limits in the console.