Ondos
Talk to us

Your archive already knows.
Ondos makes it answer.

Evidence-first intelligence infrastructure for specialist publishers in energy, shipping, and finance.

See the architecture

The producer cut October loadings to 1.8mn b/d after the pipeline outage, its lowest level in three years.

Scroll  
01 · THE PROBLEM

Generative search guesses. Your subscribers can’t afford guesses.

A specialist publisher’s entire business is being right and being citable. That is the product. It took decades of expert reporting to build, and it does not survive being approximated.

A RAG pipeline blends and paraphrases. It retrieves chunks, hands them to a model, and the model writes something plausible. The moment generation happens, provenance dies — there is no longer a sentence you can stand behind.

And an archive pumped into a vector store is an asset quietly leaking value: indexed for someone else’s answers, citable by no one, owned in name only.

02 · THE INVERSION

Most systems generate an answer, then hope it’s grounded. We invert the order.

RAG — retrieve then generate
query“how did loadings respond?”
embeddingsnearest-vector lookup
retrieved chunkspassages, ranked by similarity
generationmodel writes prose⚠ provenance dies here
answerplausible · unverifiable
Ondos — extract then return
archivedecades of expert reporting
extraction at ingestionfive passes · once, up front
structured layersentities · claims · series · edges
human adjudicationeditors confirm causal links
query returnsrecords · with provenance
NOTE
Extraction happens once, at ingestion. Retrieval returns evidence, not prose.
04 · PROOF

Our anchor publisher, ingested in full.

No logos. No testimonials at launch. The specificity is the proof.

60+ yr
weekly intelligence archive
10,000s
articles fully ingested
5
extraction passes per article
11
schema layers · L0–L10
1
sovereign corpus · owned by the publisher
05 · FEDERATION

One platform.
Sovereign corpora.

Each publisher’s corpus stays its own. Publishers keep ownership, branding, and the subscriber relationship — nothing is pooled into a common model.

Ondos provides the shared infrastructure: ingestion, the schema, and the products that sit on top. Federation, not aggregation.

Cross-corpus answers exist only where both publishers have agreed to them. What flows between nodes is a licensed query — never the underlying data.

ONDOS — — licensed query, not data — —
06 ·  THE MOAT

Your archive is the moat. Build on it.

Talk to us