← Back to the blogOpen source

Why We Open-Sourced Our Engine

Lyriks is open source under the AGPL and installs with one command. A short case for why a tool that verifies your code should be open to verification itself.

August 27, 2026 · 4 min read

We build a tool that reads your code, your tickets and your documentation, extracts the spec, and then checks every change against what you decided. It handles about the most sensitive material a software company owns. Asking you to trust it blindly would contradict the very thing it exists to do.

That’s the short version of why the engine is open source, and why there is a free Community Edition anyone can install with a single command.

The longer version starts with how teams feel about the code they already receive. Sonar’s State of Code Developer Survey (January 2026, more than 1,100 developers) found that 96% of developers don’t fully trust AI-generated code. A team that skeptical of its own pull requests is right to be at least as skeptical of the tools it installs. We would be. So the code is public. Your team can read it, audit it, see what it does with your data and confirm it opens no outbound connections.

Reading the license

We picked the AGPL deliberately. It keeps the code open even when someone uses it to provide a service, which closes the familiar route by which an open project slowly turns into a hosted black box.

For you, that protects three things. The engine you audit today stays inspectable in later versions. You can run it without us if you ever want to. And improvements stay shared instead of disappearing into a proprietary fork.

The license cuts both ways, and we chose it knowing that. Anyone can read our mistakes. Anyone can see how the engine decides that a rule is broken, and argue with it if they think it is wrong. For a verification tool, that kind of scrutiny is the point.

It also sets the terms of our relationship in public. You know now what adopting the tool commits you to, and what we can’t quietly take back later.

None of this stops us from being a company. The split is simple. The engine and the Community Edition are open and free. Paid plans (Solo Builder, Team and Enterprise) serve teams that want more, and the details are on the Pricing page. You never pay just to find out whether the tool fits.

OPEN · AGPL · FREEThe engineThe Community EditionInspect it, install it in one command,run it on your own code.PAID PLANSTeamEnterpriseFor teams of every size.
The business model: the engine and the Community Edition are open and free; paid plans serve teams of different sizes.

Proof over promises

Sovereignty claims are easy to write in a brochure. One-command install, on-premise or air-gapped, no outbound connections, your own LLM and tokens including a local or sovereign model, IP owned outright by the vendor. Any vendor could print that list.

Open code is what lets a CISO check it instead of taking our word for it. Whether anything leaves the building is something you can read in the source.

The same goes for your project memory. The spec, the documentation and the links between them live in a graph database on your side, exportable in full with their traceability. Models will keep changing, and that memory stays yours through every one of them.

IN A BROCHUREOne-command installtrust usOn-premise or air-gappedtrust usNo outbound connectionstrust usYour own LLM and tokenstrust usIN THE OPEN SOURCEOne-command installread itOn-premise or air-gappedread itNo outbound connectionsread itYour own LLM and tokensread itOpen code lets a CISO check it instead of taking our word for it.
The same sovereignty claims, printed in a brochure or readable in the source: open code lets a CISO check them.

Trying it on your own code

The Community Edition runs on your machine against your repository. Point it at a payments module and a rule like “a customer is refunded within 30 days” comes back, linked to refund.py, the Payments API and the refunds table. Change the module with your coding assistant and each rule gets checked again, with additions, omissions and regressions flagged.

The engine doing that check is deterministic, built on graph rewriting and developed in-house. Because it is open, you can trace any verdict back to the code it read and the rule it compared against. There is no hidden step, and no model whose opinion shifts between runs.

A good way to judge it is to pick a module whose rules your team already knows well. If the extracted spec matches what you would have drawn on a whiteboard, you have learned something useful. If it surfaces a rule nobody remembers writing, you have learned something more interesting. Either way the verdict comes from your own code, on your own machine, and nothing about the experiment depends on us.

Lyriks has been downloaded more than 8,000 times since May 2026, each time without a sales call standing in the way. An open tool is also an exposed one, and we’d much rather hear about a wrong note from a user’s issue than find it ourselves six months later.

Install the Community Edition, or book a demo to see the paid plans on your own code.

Keep your software in tune

Start with one module. Lyriks recovers its spec, then verifies every change. All it needs is your repository.