Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
We are on a mission to compress what the world already knows into reusable patterns, and to discover new ones, fast, in partnership with AI. Every repo, paper, and model holds patterns worth reusing. Gerolamo finds them, names them, and puts them to work. Here is how every piece fits together.
The premise is simple. The same good ideas are reinvented constantly, scattered across millions of repositories and papers in slightly different words. That duplication is waste. Gerolamo reduces it. It reads the corpus, distills each project down to the mechanisms that make it work, and names each mechanism once. A technique proven across thirty projects becomes a single reusable pattern, and those patterns are the parts you and your AI assemble into real, working things.
The whole system, in four nouns
Sources, then Patterns, then Concepts, then Enhancements.
Sources are the real projects we index. Patterns are the reusable mechanisms distilled from them. Concepts are designs grounded in those patterns, and Enhancements are the installable capabilities composed from them. Everything below is one of those four, or a tool for moving between them.
The evidence base. Every repo, paper, and model we index, searchable as one corpus.
Gerolamo indexes real projects from across the open source world, GitHub repositories, arXiv papers, and Hugging Face models. We call these sources, and they are the ground truth everything else is built on. You search the whole corpus at once. Semantic search matches on meaning, so related work surfaces even when each project describes it differently, and keyword search handles exact terms.
A robotics engineer searches 'indoor navigation without GPS' and finds a repo, a paper, and a model that all attack the same problem, even though each one names it differently.
The reusable mechanism inside a source, named once and proven by how often it recurs.
A pattern is the smallest useful unit of know-how, one mechanism that does one job, written plainly as what it takes in and what it produces. Gerolamo reads each source, extracts these mechanisms, and collapses the same pattern found across many projects into a single entry. A technique that appears in thirty projects reads as one proven pattern, not thirty near-duplicates, and you can see exactly which projects it came from. Patterns are the parts you build with.
Instead of reading ten retrieval libraries to learn how reranking works, you open one pattern, see precisely what it takes in and produces, and see the real projects it was found in.
The deliverable. A working capability you install, composed from proven patterns.
An enhancement is the thing you actually adopt, a capability composed from patterns and ready to apply. A build spec, a skill, an MCP server, a configuration, a pipeline. Each one carries a clear trail back to the patterns and sources it draws on, so you can judge why it works before you adopt it. Browse what others have published, pull one into your project, or compose your own and publish it, for free or for a price.
You need a documentation search server for your agent. You find a published enhancement that composes the right patterns, pull it in, and it works, with a clear trail back to the projects it came from.
The bench where you assemble patterns and sources into something new.
The workspace is where composition happens. Collect the patterns and sources you want to build from, choose what to produce, and the engine assembles them, drawing on the pieces that fit the target and telling you which it used and which it set aside. Your own AI does the generation, so composing is free. Build mode returns a spec with architecture and steps. Other modes produce integration briefs, comparisons, and research.
You gather five patterns for rate-limited embedding, hybrid search, and clean extraction, hit compose, and get a spec you can hand straight to your coding agent.
A design grounded in real work, before the thing exists.
A concept is a design for something that does not exist yet, assembled from real sources and patterns. You describe what you want to build, and Gerolamo keeps a precise trail back to everything it draws on. It is the blueprint, not the build. When you ship the real thing, link it back, and the concept becomes a realized project with its lineage intact.
A researcher sketches a degradation-resilient navigation system as a concept, fusing ideas from four real projects. Months later they build it, link it back, and the lineage stays intact.
The macro view. How entire fields move, not just single projects.
Analytics moves up a level, from individual projects to whole technology domains. It shows which areas are accelerating, which are saturated, and where momentum is concentrating, across dozens of tracked fields, with trends over time. Use it to read a field before you commit to it.
A strategy team checks a domain and sees it is mostly educational projects, with only a handful of serious infrastructure plays, and those few are the ones gaining momentum.
Give your assistant direct, programmatic access to the whole library.
Connect any MCP-compatible assistant to Gerolamo in a few minutes. From then on it can search sources, browse patterns, compose enhancements, and pull them into your projects on your behalf. You supply the assistant, Gerolamo supplies the building blocks and the reasoning over them. The MCP server is open source.
You tell your coding agent to research a space and draft something new. It searches, finds the right patterns, composes an enhancement, and starts building, on its own.
Reusable procedures that drive your agent through a whole task.
A workflow is a documented procedure, a sequence of steps that calls the right tools in the right order to complete a task end to end. Hand one to your assistant and it runs the whole thing, with no prompt engineering required. Workflows encode the method, so you do not have to rediscover it each time.
An engineer grabs a stack-audit workflow, pastes it into their agent, and it checks their dependencies, finds the weak ones, and proposes replacements.
Save what matters and watch how it changes.
Bookmark anything for quick access. The watchlist tracks how the things you follow change over time, what broke out, what is still growing, and what is fading, so you notice real movement without checking by hand.
An analyst bookmarks eight projects in a space. A week later the watchlist shows one jumped from fifty to four hundred stars, and two are fading.
A standing query that tells you when something new fits.
Define a query once and let it run on a schedule. When a new source appears that matches what you care about, you hear about it first. It turns search from something you repeat into something that watches for you.
A researcher sets an alert for a narrow topic. Two weeks later a new project appears that fits, and they hear about it before anyone else.
The library, reordered around what you actually work on.
Tell Gerolamo the fields, source types, and kinds of work you care about, and it tunes your results to surface those first. It is a lens, not a wall. Turn it off any time to see the full, unfiltered library.
An engineer sets their interests to robotics and autonomous agents. Now their searches lead with those fields and filter out the noise.