AI agents need version control beyond text
Changes AI agents make need to be reviewable by humans.
For code, Git solves this:
- Reviewable diffs: What exactly did the agent change?
- Human-in-the-loop: Review, then merge or reject.
- Rollback changes: Undo mistakes instantly.
But agents modify binary files too. And Git can't diff them.
Introducing Lix
Lix is a universal version control system that can diff any file format (.xlsx, .pdf, .docx, etc).
Unlike Git's line-based diffs, Lix understands file structure. Lix sees price: 10 → 12 or cell B4: pending → shipped, not "line 4 changed" or "binary files differ".
- Reviewable diffs: See exactly what an agent changed in any file format.
- Human-in-the-loop: Agents propose, humans approve.
- Safe rollback: Undo mistakes instantly.
Excel file example
An AI agent updates an order status in orders.xlsx.
Before:
| order_id | product | status |
| -------- | -------- | -------- |
| 1001 | Widget A | shipped |
| 1002 | Widget B | pending |
After:
| order_id | product | status |
| -------- | -------- | -------- |
| 1001 | Widget A | shipped |
| 1002 | Widget B | shipped |
Git sees:
-Binary files differ
Lix sees:
order_id 1002 status:
- pending
+ shipped
JSON file example
Even for structured text file formats like .json lix is tracking semantics rather than line by line diffs.
Before:
{"theme":"light","notifications":true,"language":"en"}
After:
{"theme":"dark","notifications":true,"language":"en"}
Git sees:
-{"theme":"light","notifications":true,"language":"en"}
+{"theme":"dark","notifications":true,"language":"en"}
Lix sees:
property theme:
- light
+ dark
How does Lix work?
Lix adds a version control system on top of SQL databases that let's you query virtual tables like file, file_history, etc. via plain SQL. These table's are version controlled.
Why this matters:
- Lix doesn't reinvent databases — durability, ACID, and corruption recovery are handled by battle-tested SQL databases.
- Full SQL support — query your version control system with the same SQL.
- Can runs in your existing database — no separate storage layer to manage.
┌─────────────────────────────────────────────────┐
│ Lix │
│ (version control system) │
│ │
│ ┌────────────┐ ┌──────────┐ ┌─────────┐ ┌─────┐ │
│ │ Filesystem │ │ Branches │ │ History │ │ ... │ │
│ └────────────┘ └──────────┘ └─────────┘ └─────┘ │
└────────────────────────┬────────────────────────┘
│
▼
┌─────────────────────────────────────────────────┐
│ SQL database │
└─────────────────────────────────────────────────┘
Read more about Lix architecture →
Why did we build lix?
Lix was developed alongside inlang, open-source localization infrastructure.
We had to develop a new version control system that addressed git's limitations inlang ran into, see (see "Git is unsuited for applications"). The result is Lix, now at over 90k weekly downloads on NPM.
Getting started
Python ·
npm install @lix-js/sdk
import { openLix } from "@lix-js/sdk";
const lix = await openLix({
environment: new InMemorySQLite()
});
await lix.db.insertInto("file").values({ path: "/hello.txt", data: ... }).execute();
const diff = selectWorkingDiff({ lix })
What's next
The next version of Lix will be a refactor to be purely "preprocessor" based. This enables:
- Fast writes (RFC 001)
- Any SQL database (SQLite, Postgres, Turso, MySQL)
- SDKs for Python, Rust, Go (RFC 002)
┌────────────────┐
SELECT * FROM ... │ Lix Engine │ SELECT * FROM ...
───────────────────▶ │ (Rust) │ ───────────────────▶ Database
└────────────────┘