Orumilos
Closed Beta

Orumilos

Context Engine

From raw documents to structured, auditable intelligence. Not another RAG wrapper.

Extract typed claims, entities, and relations from any source — with full provenance. Built for developers whose AI applications need to prove where answers come from.

Facts extracted
PDF · DOCX · TXT · URLs
Full provenance tracing
Compare vs. standard RAG

RAG gets you to demo. Then what?

Can't debug failures

Bad chunks? Wrong retrieval? Hallucination? You can't tell.

No provenance

Your AI gave an answer. Can you prove where it came from?

Entity chaos

Same company, 10 different strings. No resolution.

Stale data

Facts expire. Contradictions pile up. Nothing versioned.

What Orumilos produces from your documents

Your document

Unstructured text. No types. No scores. No way to verify.

Orumilos output
factSystem must support 10k concurrent users94%
decisionPostgreSQL selected as primary datastore88%
requirementAll API responses under 200ms p9991%
entityPostgreSQL= Postgres, PG
relationPostgreSQL addressed_by caching layer

Typed claims with confidence scores. Full provenance trace.

How Orumilos is different

Provenance-First

Every claim traces to its source. A write-time invariant — not optional.

Claim Chunk Artifact

Typed Primitives

Structured claims with confidence scores. Queryable without LLM inference.

fact decision requirement objective 94%

Entity Resolution

Aliases merged. One canonical truth per real-world thing.

"Acme Inc" "ACME" "acme corp"
Acme Inc

Versioned Artifacts

Sources update. Downstream packs marked stale. Full reproducibility.

v1
stale
v2
v3
current

EU AI Act enforcement begins August 2026

AI systems must demonstrate provenance and traceability. Fines up to €35M or 7% of revenue. Orumilos gives you the audit trail from day one.

Invest at ingestion, save at query time

Standard RAG

queries →
  • • LLM inference per query
  • • Linear cost scaling
  • • No structure or versioning

Orumilos

ingestion queries →
  • • Extract once, query storage
  • • Cost flattens after ingestion
  • • Full provenance included

The more you query, the more the model pays for itself.

Built for AI applications that need trust

Meeting Intelligence

Extract decisions, action items, and commitments. Every item links to the exact speaker turn.

Research & Analysis

Process technical docs, legal briefs, or reports. Query structured claims instead of re-reading.

Compliance & Audit

Every AI output traces to its source. Provenance-ready for regulatory requirements like the EU AI Act.

Developer-first interface

orumilos-cli
# ingest a web source
orumilos run pipeline web --url "https://example.com/report.pdf"
 
# inspect extracted context
orumilos inspect claims --workspace 01JQXR...
✓ 47 claims · 12 entities · 8 relations
 
# assemble a context pack
orumilos pack generate --workspace 01JQXR... --query "Orumilos"
★ Pack ready — 5 cited claims, 3 sources
python
from orumilos import Client
 
client = Client(api_key="...")
job = client.ingest("report.pdf")
claims = client.claims(artifact=job.artifact_id)
 
for claim in claims:
  print(f"[{claim.type}] {claim.text}")
  print(f" Source: {claim.evidence.chunk_text[:80]}")

Common questions

Standard RAG stores raw text chunks and retrieves by similarity. Orumilos extracts typed, scored primitives — claims, entities, relations — and traces every one back to its source. You get structured intelligence, not a bag of text.

Every extracted fact links to the exact passage it came from — artifact, chunk, and character offsets. You can verify any claim against the original document. No black-box answers.

A structured briefing assembled from your extracted intelligence, with inline citations. Unlike RAG retrieval, every statement in a pack is backed by a typed, confidence-scored claim.

PDF, DOCX, TXT files and web pages via URL. Transcript formats with speaker turns. More connectors coming.

The kernel (V1) is complete. API, auth, async jobs, and production adapters are built. We're onboarding early teams in closed beta.

Ready to structure your knowledge?

Start extracting structured, traceable intelligence from your documents.