@diptanu Just use Datafusion.
We did use SQLite as a parser and planner. One workaround after another. Datafusion is so much better because it gives full control above the storage layout.
@enrique_goudet@lixCCS@lix lix stores files and version control info in your existing postgres. so you can use postgres for your normal SQL database stuff and get a version controlled filesystem on top of your postgres
Announcing @lixCCS v0.6: An embeddable version control system.
Import version control as a library, bring your own backend.
```js
import { openLix } from "@lix-js/sdk"
const lix = await openLix({})
lix.fs.writeFile("/hello.md", "Hello World")
```
AI agents are creating an explosion in version control demand. AI agents need isolated workspaces, checkpoints, branches, reviewable changes, rollback, and durable state.
Today, teams try to solve this by wrapping Git in their infrastructure.
But Git was not designed to be embedded. You end up managing repositories, locks, packfiles, GC, LFS, process calls, protocol servers, and cross-system transaction coordination around a tool that expects to live outside your application and infrastructure.
Lix solves this with an embeddable version control system you import as a library, backed by your own backend:
- Bring your own backend: SQLite, Postgres, S3 object storage, Cloudflare storage, etc.
- Handles any file format (.docx, .pdf, etc.)
- Large-file support built in, without a separate LFS layer
- ACID transactions across state, blobs, and history
- SQL as a query interface for agents
What you can build with Lix:
- AI agent filesystems with branches, checkpoints, and rollback
- Version-control extensions for databases like Postgres or SQLite
- Data lakes with commits, diffs, rollback, and large-file support
- and more...
More releases will follow that expand the use cases, but v0.6 is the first usable release of the core idea: a version control system you can embed.
Lix April 2026 update:
- Benchmarking exposed that SQLite gives too little control over Lix's versioned storage model to keep improving incrementally.
- Decision: move query execution to DataFusion while keeping SQLite as a possible physical storage backend.
- May goal: Release `v0.6` MVP with focus on CRUD with branching and merging on the optimized semantic write path that the file API will use next.
---
You can already try out lix v0.6 via NPM:
```js
npm i @lix-js/[email protected]
```
The next step is to finalize the physical storage layout. The runtime can be optimized incrementally in follow up releases.
Lix v0.6-preview.5 shipped with an ACID transaction API.
That leads to up to 17.6x faster writes, and more importantly grouped state updates are now atomic across commits.
Lix v0.6-preview.5 shipped with an ACID transaction API.
That leads to up to 17.6x faster writes, and more importantly grouped state updates are now atomic across commits.
Biggest red flag I hear from early stage founders?
โWeโre working on self-serve onboardingโ
Your product currently requires you to talk to every new customer?
Thatโs a good thing.
Opened a PR to speed up CSV imports in Agent CRM by @enrique_goudet.
The test fixture imported in 1.3 seconds instead of 64 seconds (โ47x faster)
https://t.co/EdoDlyN9FX