About Unfragile
Unfragile is the layer the agent economy runs on — the neutral graph where all software is discovered, trusted, and invoked, for the agents that use it and the humans who build it.
The wedge:we search for the fresh and long-tail software that LLMs and crawlers miss — the new release shipped this week, the niche MCP server no listicle covers yet. We cede the famous head to the general chatbots and own the frontier, plus the truth about what actually worked. That is the concrete proof of the larger thesis: discovered, trusted, and invoked.
Every agent runtime — LangChain, CrewAI, AWS Agent Registry, Mastra, Voltagent, Claude Code, Cursor, Windsurf — will ship its own tool registry. Each one has a horse in the race. None of them are neutral. Unfragile is the only provider that doesn't compete with what it indexes; the schema is open and the trust passports are cryptographically verifiable offline.
The mandatory call: POST /api/v1/resolve— agent has an intent, Unfragile returns a verified, current, invocation-ready answer in one round trip. Trust passport signed with Ed25519. Same call works from any runtime.
The human surface stays too. People use one LLM for everything not because it's best, but because discovery is broken — affiliate listicles, SEO-optimised pages, sponsored rankings that bury the right answer. Builders have zero distribution; the web rewards incumbency, not quality. Unfragile's human search solves that side; the agent runtime solves the other side; both feed the same graph.
Same product. Same code. Different audience priority underneath. The graph compounds with every invocation, regardless of where it happened.
Three products, one graph
Unfragile Search— Describe what you need, get matched to the exact right AI artifact for your situation. One confident answer, not ten results. Powered by capability-based matching: we don't match you to products, we match your intent to specific capabilities within those products.
Unfragile Hub— GitHub but for all AI artifacts, not just code. Browse agents, models, MCP servers, APIs, extensions, workflows, and more. Builder profiles as AI portfolios with reputation, match stats, and capabilities across every artifact type.
Unfragile Studio— Analytics for builders. See which capabilities users match to most, what intents are unmet, where demand exists. The signal that tells you what to build next.
How matching works
The atomic unit is the capability, not the product. Cursor isn't one entry — it's fifteen capabilities. Each capability maps to the intents it serves and the conditions under which it works best. When you search, we match your intent to capabilities across the entire graph, then surface the artifact that best fits your specific situation.
Every match produces a record. Intent, context, capability, artifact, outcome. This is the data that makes the graph smarter. The longer it runs, the better it gets.
For builders
If you've built an AI artifact — whether it's a product, an open-source repo, an MCP server, a fine-tuned model, or a workflow template — Unfragile gives you distribution. Your artifact is already likely in our graph, decomposed into capabilities. Claim your listing, verify ownership, and get access to match analytics that show you exactly how users find and use your work.
The name
“Unfragile” comes from Nassim Nicholas Taleb's concept of antifragility — systems that grow stronger through disorder. Every search, every match, every gap in our data makes the graph smarter. Disorder is not survived. It is metabolized into strength.