Skelf Research: The Independent Lab That Ships Research as Code

The four kinds of AI research organisations

The AI tooling ecosystem in 2026 has four distinct kinds of research organisations. The way they publish, the way they fund, and the way they measure success are all different. Skelf Research sits in a category most people don’t have a name for.

Kind	Examples	Output	Funding	Audience
Big Tech AI labs	DeepMind, Anthropic, OpenAI, Meta AI	Papers, products	Corporate	Shareholders + academia
AI startups	Mistral, Cohere, Anyscale	Products (sometimes papers)	Venture	Customers + investors
Individual researchers	Andrej Karpathy, Simon Willison, etc.	Posts, code, courses	Patreon, courses	Followers
Independent research labs	Skelf Research, MILA spinoffs, some academic groups	Open-source + papers	Grants, founders, partnerships	Community

Skelf Research is the fourth kind. We are not a Big Tech lab (we don’t have a product, shareholders, or a research budget from a corporation). We are not an AI startup (we don’t raise venture capital, we don’t sell a SaaS). We are not an individual researcher (we are a lab with multiple projects, a methodology, and a long-term roadmap).

We are an independent research lab that ships research as code.

What that means in practice

The operational rules we follow:

No products. We don’t sell a SaaS, license a proprietary engine, or take VC funding. The output is the public repositories, not a hosted service.
Hypotheses as software. Every research question is encoded in a runnable, testable, peer-reviewable repository. The codebase is the proof.
MIT or GPL-3.0. Every project is open-source. We use copyleft (GPL-3.0) for the projects that benefit most from it (memista, liath, embedcache, polymathy, liath-rs) and permissive (MIT) for the rest.
Multi-language by design. Rust, Go, Zig, Python, Dart, Lua, JavaScript, TypeScript. The thesis is language-agnostic; we choose the language for falsifiability, not fashion.
Geographic and regulatory positioning. UK base. EU AI Act, GDPR, SOX, HIPAA, compliance posture built into the architecture from day one.

Why “independent research lab” matters

The category matters because of what it enables:

Reproducibility. A Big Tech paper with a non-public implementation is a black box. A Skelf paper with a public repository is a falsifiable claim. The methodology is the same as the reproducibility movement in experimental science.
Trust. A vendor selling an LLM observability tool has a reason to overstate its capabilities. An independent lab publishing the measurements has the opposite incentive. Trust accrues to the institution that is most verifiable.
Compounding. A Big Tech AI lab’s output lives in a closed system. A Skelf project’s output lives in the public commons, where it can be forked, improved, and built upon. The compounding rate of public research is dramatically higher.
Hiring. The best AI researchers want to publish their work in the open. The independent-lab structure is the only one that allows that, while also letting them ship production-grade code.

How Skelf compares to specific reference points

Reference	What they do	How Skelf differs
DeepMind	Publishes papers, builds products, has a corporate parent	We don’t publish papers as our primary output; we publish code. No corporate parent.
OpenAI	Builds frontier models, sells API access	We don’t build frontier models. We build the infrastructure around LLMs.
LangChain	Python framework for LLM apps	We are not a framework. We are individual projects that compose with frameworks.
DSPy	Python framework for prompt optimisation	We are not a framework. Our promptel is a specification language; DSPy is a compiler.
Vercel	Hosts web apps	We don’t host anything. We publish software.
Academic research lab	Publishes papers with code, trains students	We don’t have students. We have a roadmap, and we ship production-grade code.
Open source foundations	Stewards major OSS projects (Linux, Apache)	We are a single lab, not a foundation. Our output is Skelf’s research, not community software.

What we are NOT

To be clear about what we are not:

Not a research lab in the academic sense. We don’t have PhD students, we don’t publish papers in peer-reviewed venues (yet), we don’t accept grant funding (yet).
Not a company in the venture sense. We don’t raise capital, we don’t have a board, we don’t have an exit strategy.
Not a foundation in the Linux sense. We don’t steward third-party projects; we are the source of our own.
Not a one-person operation. We are a lab with a methodology, a roadmap, and multiple projects in production.

The moat

The moat is the methodology, not the projects. A competitor can copy a project (mullama, promptel, zviz). What they cannot copy is the culture of hypotheses as software: every research question becomes a runnable, testable, peer-reviewable artefact.

This compounds. Every year we have more projects, more articles, more glossary entries, more comparison content, more discoverability in AI search. The moat is the corpus, not any individual piece.

What’s coming

Roadmap (2026 H2):

3-5 new repositories in the existing pillars
12+ new articles (we currently have 15+)
Series landing pages
Per-pillar landing pages
Expanded glossary to 50+ terms
More “vs” comparison articles
More personas covered (engineering leaders, domain practitioners)

The independent-lab model is replicable. Anyone with the methodology can start one. But the compounding advantage of an early start + a public corpus is hard to close.

How to engage

Use the projects. pip install / cargo add / npm install / pub add as appropriate.
Read the articles. They’re peer-reviewable. Cite them if they’re useful.
File issues, send PRs. All 25 repositories accept contributions.
Hire us for research collaborations. We work with academic and industry partners on research-driven software. contact@skelfresearch.com.
Fund the lab. We are funded by the founders and research partnerships. If you are a foundation or research organisation interested in supporting specific projects, get in touch.