---
post_type: "project"
title: "Adapt"
blurb: "Open-source Node.js library for building memory systems that learn and evolve over time, instead of just storing and retrieving data."
org: unbody-lab
tags:
    - A.I. software
    - Open source
    - Machine learning
    - Adaptive memory
year:
    - 2026
current: true
link: "https://adapt.unbody.io"
stack:
    - TypeScript
    - Node.js
---
[Adapt](https://adapt.unbody.io) is an open-source Node.js library for building memory systems that learn and evolve over time, rather than just storing and retrieving data.

Most memory systems today fail in two ways: they can't reason across time, and they can't adapt when data changes shape. They memorize instead of learning — a fixed structure trying to hold non-deterministic data. Adapt is my attempt to fix that.

It runs on two components. A **Brain** orchestrates the system — routing incoming data, synthesizing answers, and monitoring its own coverage. **Neurons** are specialists that each learn one slice of the domain: they observe new data, score it by relevance, and compress it into understanding instead of hoarding raw records. When the Brain detects gaps or redundancy, it reconfigures itself — splitting overloaded neurons, merging similar ones, or creating new ones for patterns it was ignoring. No schema migrations, no manual reconfiguration.

You steer it with three operations: `inject()` to feed it data, `signal()` to nudge structural change, and `adjust()` to tune a single neuron. Querying is separate from learning — questions consult accumulated neuron knowledge and come back with relevance and confidence scores.

- Site: [adapt.unbody.io](https://adapt.unbody.io)
- Docs: [adapt.unbody.io/docs](https://adapt.unbody.io/docs) — and the [core concepts](https://adapt.unbody.io/docs/concepts)
- Source: [github.com/unbody-io/adapt](https://github.com/unbody-io/adapt)
- Background: [Why I built Adapt](https://unbody.io/blog/adapt)