The agentic last mile
HydrAI is one piece of a bigger idea: that AI gets safe, accurate, and cost-effective when data carries its own meaning. This page is about HydrAI's piece — extending that meaning to the surface an agent acts through. It stands on its own; the full thesis lives elsewhere, and we leave breadcrumbs.
The foundation, briefly
An LLM has no mechanism for truth — only probability. That gap isn't closed by a bigger model; it's closed by better data. Semantically grounded data — JSON-LD layered over the APIs you already have, additive and non-breaking — turns "magic strings" the model has to guess at into terms with identity it can actually rely on. Give a model facts instead of a puzzle, and its answers become discoverable, justifiable, and cheap.
That is the foundation HydrAI stands on, and it is a deep story in its own right — the semantic layer, the chains of justification, integration that stops being a problem. HydrAI does not retell it here.
Meaning doesn't have to stop with your data
Here is the leap HydrAI cares about. Once your systems can communicate with meaning, you can describe more than what a field means. You can describe your system's own surface for interaction — what an agent can do, where it can go, how to build a query.
Today we compensate for the absence of that with custom tools, hand-written MCP servers, and hundred-thousand-token prompts that memorize an API in advance. Imagine a browser that had to hard-code every website's map before it could load a page. That is how we build agents now.
It doesn't have to be that way, because describing an interaction surface is also a semantics problem — and Hydra already solves it. Hydra is a lightweight JSON-LD vocabulary for an API's affordances, carried in the responses themselves. A generic agent connects with one fact — the entry-point URL — and discovers the rest. No prompt, no manifest, no custom code. The same way your browser reads a search form out of the page and simply knows how to use it.
This is a web for agents: self-describing interoperability that falls out for free once the surface means something. Try it — point the generic agent at a Hydra API and watch it navigate, filter, and act against a system it was never told about.
Hydra's stable core, extended for the age of agents
Hydra is a stable, robust core — essentially finished, because it is a foundation. But it was built for machines that navigate. An agent also reasons in examples and plans and acts, and Hydra has no vocabulary for the few things those require: a greeting to orient on, worked examples to learn from, consequence semantics to plan against.
That gap is HydrAI: a conservative superset of Hydra core — everything Hydra says, plus the agentic last mile, minted only at the gap and always additive. The Hydra Community Group is dormant, so HydrAI stewards the core unchanged and adds the missing terms on top.
The aim is to be the one-stop shop for building domain-agnostic, agent-friendly hypermedia APIs: one vocabulary that covers the whole interaction surface, so you learn it once instead of assembling a dozen. It stays domain-agnostic on purpose — HydrAI describes how to interact with an API, never what your API is about, so your own entities keep their own vocabulary and ride in on your API's context.
Standards, not walled gardens
A universal semantic layer can't come with lock-in, so HydrAI lives at a neutral namespace, on open standards, as open-source reference code. It is not a product or a platform — it is a discipline, additive by design: start with one API that describes itself, break nothing, and grow from there.