Generative AI you can prove

Ask, and get a governed answer you can prove: grounded, cited, scored, and audited.

Try:

Feed Akeel knowledge fast - import a whole standard ontology, paste your own RDF/Turtle, or add a single fact. Everything passes the reasoner + rules gate before it's accepted.

What Akeel knows

Standard vocabularies come pre-loaded; below that is your own knowledge.
standard vocabularies loaded
Your knowledge
facts stated
facts inferred
What Akeel figured out (nobody typed these):

Add a vocabulary

The standard public ontologies are already loaded. Use this to add a new or custom one - pick from the list, paste your own RDF, or fetch any URL.
Or paste RDF / Turtle:
Or import any public ontology by URL:
Imports pass the reasoner + SHACL gate — inconsistent or malformed data is rejected, not silently loaded.

🧠 Ontology thinking — your data, understood through the standard vocabularies

This is Akeel's core advantage: it doesn't just store the loaded vocabularies, it reasons across them. By aligning your data to FOAF and the W3C Org ontology, the reasoner derives facts nobody typed — chaining from your own terms into the standards.
Every line is inferred by tested code, not the LLM. Ask the Assistant "Alice: what is she classified as across all vocabularies?" and watch it answer, grounded and cited, from these derived facts.

Ask your data

Precise answers from the knowledge base - including facts reached only by reasoning.
Add a single fact (advanced)
Offline structured form - one operation at a time. Also: demo sets, and plain-English (needs a connected model).
Demo sets:
Plain English:
Akeel accepts a change only if it stays consistent and passes your data rules.

A governed pipeline is how Akeel automates work you can trust. Each request flows through five checked stages: retrieve → generate → gate → judge → record. Nothing ships unless it passes.

Configure how Akeel runs - your on-device models and an optional cloud provider. Changes apply to this running session; secrets stay in memory and are never written to disk.

Cloud model

Cloud is an external model, powered by an API or a CLI you are signed into. Pick one, then use the Cloud toggle.
Default: none — loads on startup. Saved providers persist; the API key never does.

Local (Ollama)

On-device models - private, no data leaves your machine.

Personality

Choose who answers you — Akeel (عقيل, male) or Akeela (عقيلة, female), and how they speak. This is the style channel only: it changes voice and framing, never the facts. The same question returns the same cited facts under any personality.

Your profile — so Akeel can build a relationship

Akeel adapts how it explains to you (and which facts it foregrounds) — never the facts themselves. This is local and yours: stored on your machine, editable, and deletable at any time.

Display & diagnostics

Tune the Ontology-thinking panel, and turn on request/model tracing if the UI ever feels stuck (logs go to the terminal and ~/.cairn/cairn.log).
Hiding drops very abstract but correct types (a Person is a geo:SpatialThing) from the panel only — the fact stays in the graph and in queries.