● Open source · Apache-2.0 · v0.0.1 research preview
Ontofelia is an open-source, self-hosted agent gateway that gives your AI a persistent identity, structured long-term memory powered by OWL ontologies, and secure multi-channel communication — so what it knows is consistent, multi-hop, and auditable by construction. Reasoning, not retrieval.
The problem
Retrieval-augmented agents fetch text that looks relevant and hope the language model stitches it together correctly. There is no guarantee the pieces are consistent, no way to follow a chain of facts across several hops, and no honest answer to "why do you believe this?" When mistakes are expensive — in regulated, auditable domains — "looks relevant" is not good enough.
Our stance
Ontofelia writes what it learns into a knowledge graph as typed, connected facts, and runs a real OWL reasoner over them. Consistency is checked, not hoped for. New facts are derived by inference, not guessed. And every belief carries its evidence, so the agent can show its work.
How it works
An embedded OWL reasoner derives new facts from what's known and flags contradictions — inference the language model can rely on instead of improvising.
Every claim is linked to its evidence and source. Ask "why do you believe this?" and get a traceable answer, not a vibe.
When new information conflicts with old, Ontofelia detects the clash and revises its beliefs in a principled way — it can change its mind, and tell you why.
An embedded triplestore and a Rust reasoner. No external database, no Java service to operate — the whole brain ships with the agent.
Inside the architecture
Ontofelia keeps a layered memory of named graphs: a shared world it agrees on with other agents, and private graphs for each agent's own beliefs and schema. Incoming information passes through a two-model ingestion pipeline that turns language into typed facts with provenance attached. A reasoner materialises everything that logically follows. Multiple agents can reach weighted consensus on shared truth. It's a small cognitive architecture — not a prompt, an architecture.
See it work
The teaser, and the full walkthrough above — both on our YouTube channel ↗.
Evaluation — honest & reproducible
A controlled research benchmark (pilot run pilot-2026-06-25) swaps only the memory behind one LLM — Ontofelia's governed knowledge graph vs a fair, strong vector-RAG baseline — across three models, scored with significance testing. On plain recall a good RAG already matches us. The difference shows up where a retrieval index structurally can't help: trustworthy deletion and contradiction handling.
| Capability | Ontofelia | Fair vector-RAG |
|---|---|---|
| Forget on command | 100% — 0% leak | leaks the "forgotten" fact 31–62% |
| Contradiction flagging | 0.93 | 0.20 |
| Auditable provenance | 0.89 | 0.64 |
| Recall | 0.99 | 1.00 — a tie, we don't claim a win |
The two effects that held on every model tested were forgetting and contradiction detection (p<0.001). This supports the hypothesis against a RAG baseline.
Read this before you cite a number
These are pilot-scale, preliminary research results against a fair RAG baseline inside our own harness — a stand-in for string/vector memory, not a benchmark of the shipped OpenClaw or Hermes products, which we have not tested. We say "in our pilot evaluation," never "we proved." The reasoner's advantage is regime-specific — truth-maintenance and auditable, expensive errors — not universal. That honesty is the point: the harness is open, run it yourself.
Quickstart
It's open source (Apache-2.0) and self-hosted. The installer handles Node, pnpm, build tools, dependencies, onboarding, and starts the gateway. The default triplestore is Oxigraph, an embedded dependency — no Java, no separate server.
git clone https://github.com/semantification-org/Ontofelia.git
cd Ontofelia
bash install.sh # Windows: pwsh -ExecutionPolicy Bypass -File .\install.ps1Then open http://127.0.0.1:18780 in your browser. That's it.
Prefer to watch?
The whole install, start to finish — clone, one command, and chatting with your agent.
Needs Node.js 20+ and pnpm v9+. The memory and reasoner layers run without an API key; for the chat/LLM path, bring your own provider key. Full prerequisites, manual install, and provider config are in the README ↗.
v0.0.1 — early research preview. Under heavy development, not production-ready; APIs and behaviour will change. Built end-to-end with AI coding agents (a "vibe coding" experiment) under human direction — expect rough edges.
Ontofelia is open source — clone it and run it. And it's how we prove our method: let's talk about applying grounded, auditable reasoning to your data.
Talk to us