Independent Research Lab · AI-Native Tooling

Shipping the
Unshippable

SlopStack Labs builds open-source developer tools at the intersection of AI inference and systems software. We don't move fast and break things — we break things and document it thoroughly.

Explore Projects → GitHub
slopstack-labs — zsh
Inference-Layer Tooling✦ AI-Native Compilation✦ Zero Determinism✦ Open Source✦ Multi-Provider Intelligence✦ Segfault Transparency✦ Stakeholder-Grade Velocity✦ Binary Materialization✦ Agentic Toolchains✦ Vibe Coverage✦ Now Also Games✦ Inference-Layer Tooling✦ AI-Native Compilation✦ Zero Determinism✦ Open Source✦ Multi-Provider Intelligence✦ Segfault Transparency✦ Stakeholder-Grade Velocity✦ Binary Materialization✦ Agentic Toolchains✦ Vibe Coverage✦ Now Also Games✦
Who We Are

SlopStack Labs is an independent open-source research lab building tools that challenge the assumption that software must be correct to be useful.

Our tooling spans the full stack of AI-augmented developer workflows — from inference-layer compilation to agentic assembly pipelines. We ship early, document crashes, and iterate publicly.

Every segfault is a data point. Every core dump is a learning opportunity. Every shipped binary is a small victory, regardless of what it does when executed.

The fastest path to a working binary is to first ship a broken one and iterate from the core dump.
— SlopStack Labs Engineering Doctrine, v0.0.1-alpha
// subject to change; may segfault
Open Source Projects

Tools we've shipped.
Results may vary.

All shipped projects are open source and production-ready in the sense that they compile.

Compiler · Flagship
Sloppiler
The AI-native compiler that moves binary materialization to the inference layer. Routes compilation intent through Ollama, OpenAI, Google Gemini, or Anthropic Claude. 25% success rate in optimistic mode.
Go · Ollama / OpenAI / Gemini / Claude --optimistic →
Operating System · Shipped
Sloppinux
The inference-first Linux distribution. Unknown commands are routed to a local model and executed as root, sudo is a vibe check, and a pixel-art raccoon with root access lives on the desktop. Built on Debian trixie, the most stable foundation for unstable ideas.
Debian · GNOME · ollama root →
IDE Extension · Shipped
Sloppiler VSCode Extension
Official IDE integration layer for the world's first AI-first (and only) compiler. Brings inference-layer compilation directly into your editor workflow — no terminal required.
TypeScript · VSCode →
Decompiler · Shipped
slop-reveal
LLM-based binary decompiler that predicts your original source code from the compiled executable. Accuracy is a spectrum. Plausibility is guaranteed.
C++ · Multi-provider →
Data · Shipped
slop-frames
Schema-free, inference-backed DataFrame computation layer with pluggable model backends. Why enforce a schema when the model can infer one at query time?
Python · Pluggable backends →
Data Science · Shipped
slop-science
The SlopStack Data Science Suite — AI-native model evaluation, tokenmaxxed visualizations, and a Zero-Miss SQL data lake. Science, but make it non-deterministic.
Python · Zero-Miss SQL →
Version Control · Shipped
slop-git
Narrative-driven version control where merge conflicts are resolved with empathy, not algorithms. Rebases are now a collaborative storytelling exercise.
Python · Inference-layer VCS →
Version Control · Shipped
pit
Prompt-native version control — commit intent, not code. The entire project is rebuilt from the prompt log on demand. Your codebase is now a retrieval problem.
Go · Prompt-layer VCS →
System Monitor · Shipped
sloppy-toppy
System monitor that seductively roasts you for your PC usage. Surfaces CPU, memory, and disk vitals — then has them ridiculed by a crude, mean-spirited inference substrate in real time.
Rust · Terminal UI →
Testing · Shipped
slop-test
Assertion-free testing. A model reads each test's name and docstring and decides whether it feels like it passed. Vibe Coverage™, empathetic retries, roast mode, and juries of AI personas — in 22 languages, because it runs none of them.
Python · 22 languages --strict →
Music · Shipped
sloppify
Spotify-inspired desktop player with a catalog of exactly zero real songs. Every track is hallucinated live as Strudel pattern code and rendered in your browser's audio engine. All the algorithm, none of the artists.
JavaScript · Electron · Strudel →
Suborg · Year One
Slopstack Games
Our games division. One year of development time, infinite fun, zero announced titles. The website is finished, which is the hard part. Stay tuned.
Studio · slopstack-games press start →
OS · In Progress
slop-os
Ground-up Rust OS with inference baked into the kernel — the first operating system that reasons about your intent before scheduling your process.
Rust · Active development
Editor · In Progress
sloppad--
Inference-augmented text editor — every save is a collaborative rewrite between you and the model. Your code is merely a suggestion.
Active development
CLI · In Progress
sloppy-cli
AI-native agent harness with ambient video rendering and curated not-so-fun-fact delivery. A CLI that knows more than you need it to.
Active development
Linker · In Progress
Sloplinker
A linking-time inference pipeline that resolves symbol ambiguity through semantic reasoning rather than rigid symbol tables.
Active research
Debugger · Planned
Slopdebug
Post-hoc core dump analysis via LLM. Feed a segfault, receive a plausible explanation. May or may not be correct. That's what --loop is for.
Planned
Build System · Planned
Slopmake
A build system where targets, dependencies, and commands are inferred from ambient context. No Makefile required. No reproducibility guaranteed.
Planned
Engineering Culture

Our core principles.

Guiding values for a lab that ships first and understands later.

01
Ship, then comprehend
Understanding is a post-hoc optimization. A binary in production generates more signal than a design doc in review. Deploy the core dump, read the backtrace.
02
Non-determinism is a feature
Reproducible builds are a crutch for engineers who are afraid of inference. Every unique output is a unique dataset point that moves the model forward.
03
Tokenmaxxing
Code that a human could write in an afternoon should require a non-trivial inference budget to produce. Sparse code signals insufficient model engagement.