Your AI should get smarter every time it runs.

Fractal gives your agents and models recursive self-improvement and continual learning with just a few API calls. Performance compounds over time — automatically.


/ Why Fractal

01

Recursive self‑improvement

Your models learn from their own outputs. Each cycle refines reasoning, reduces errors, and sharpens decision‑making without manual retraining.

02

Continual learning loops

Performance improves the longer your system runs. Our learning loops adapt to new data in real time, so your AI never goes stale.

03

A few API calls

No infrastructure overhaul. Integrate Fractal into your existing stack with minimal code. Start improving your agents in minutes, not months.


/ How it works

  1. 01

    Connect your agent or model

    Point Fractal at your existing pipeline. Works with any LLM provider, framework, or custom agent architecture.

  2. 02

    Define your learning objective

    Set the metrics that matter — accuracy, latency, cost, user satisfaction — and Fractal optimizes toward them.

  3. 03

    Watch performance compound

    Every interaction feeds the learning loop. Your system gets measurably better with each cycle — and the gains accelerate.

index.ts
import { Fractal } from "fractal-sdk";

const fractal = new Fractal({
  apiKey: process.env.FRACTAL_API_KEY,
});

// Start a learning loop
const loop = await fractal.loops.create({
  agent: "my-agent",
  objective: "accuracy",
});

// Your agent improves automatically
const result = await loop.run(input);

/ Built for

Agent developers

Give your autonomous agents the ability to learn from every task they complete.

ML teams

Close the feedback loop between production performance and model improvement.

AI-native startups

Ship self-improving products without building custom training infrastructure.

Enterprise AI

Deploy sovereign AI that learns on your data and stays within your environment.

Make your AI better, every single run.

Get early access to Fractal and start building self-improving systems today.