Skip to main content

Increase your RAG accuracy by joining Orpheam's Pilot Program.Request your demo here

Orpheam

Where knowledge finds its harmony

Fly through the knowledge graph

PDF · 248p
export.csv
Entity
Relation
Constraint
role: ops
scope: org
cite · 96%
reasoning path
signed
who · when
replay
01 / 05

Your documents

PDFs, procedures, regulations and exports enter as they are — no reshaping.

Procedure_2024.pdf248p
EU_Reg_881.pdf61p
vendors.csv3.2k
clauseentitydateamount
12 sources · 4 fragments
02 / 05

Knowledge graph

Entities, relations and constraints become a validated graph on Memgraph.

multi-hop · ASP-repaired
03 / 05

Per-query ACL

Permissions are enforced before generation — the model only sees what the user can.

DocsGraphPII
Analyst
Ops
Guest
org → entity scope
04 / 05

Sourced answer

Native citations, per-fragment confidence and a replayable reasoning path.

“Export is permitted under clause 4.2 when the vendor is EU-based.”

123
precision
96%
recall
88%
fresh
81%
96 · 88 · 81% confidence
05 / 05

Audit trail

Who asked what, when, on which source — exportable, regulator-grade.

09:12ops@query · 12 src
09:12systemACL pass
09:13auditsigned · stored
signed · replayable

Create.Connect.Compose.

+30%accuracy
<20msgraph traversal
EUsovereign
Start creatingWatch demo

Trusted by document-heavy teams in regulated industries

How it works

01 / 05

  1. 01

    Your documents, as they are

    PDFs, procedures, regulations, exports, APIs. No reshaping — Orpheam meets your sources where they live.

    • Tables, schemas & reading order preserved
    • Connectors for files, APIs and databases

  2. 02

    Entities & relations, surfaced

    NLP reads each document and proposes entities, relations and constraints for your experts to approve.

    • Domain ontology in your business language
    • Expert-in-the-loop validation

  3. 03

    A validated knowledge graph

    Approved facts become a graph on Neo4j. Symbolic repair (ASP) removes incoherence before retrieval.

    • Cross-document relations, not isolated chunks
    • Incremental patch — no full reindex

  4. 04

    Permissions travel with the query

    Per-query ACL is enforced before generation. The model only ever traverses what the user may see.

    • Org → workspace → project → entity scope
    • Full, exportable access trail

  5. 05

    An answer you can open

    The response ships with its proofs — exact fragment, page, path and a confidence score per source.

    • Native citations, replayable reasoning
    • Regulator-ready audit export
orpheam · stage 1/5INGEST

Sources

PDF
DOCX
CSV
API
XLSX
JSON
Secure ingest · EU
regulation.pdf

Extracted

EntityDORA
Entitycritical provider
Relationapplies_to
ConstraintArt. 28-31

Policy

rolecompliance.read
scopews · compliance-eu
effectevaluating
allowed denied
Orpheam answer composing

Cited sources

Reg. (EU) 2022/2554 — DORA0%
ESAs — ICT Third-Party Risk0%
Internal Compliance memo0%
24 documents ingested

Beyond vector RAG

Not a better RAG — a new paradigm

Vector search retrieves look-alike text. GraphRAG reasons over a living knowledge graph — held in memory on Memgraph — so answers are related, real-time and verifiable.

Yesterday

Vector RAG

  • Text sliced into isolated chunks
  • Retrieval by similarity, not meaning
  • Plausible answers, approximate sources
  • Permissions filtered after generation
  • Re-index the whole corpus on change
  • Breaks on cross-document relations
New paradigm

GraphRAG on Memgraph

  • Entities & relations in a real graph
  • Multi-hop traversal across documents
  • Native citations, per-fragment scores
  • Per-query ACL enforced before generation
  • Real-time incremental patch — no reindex
  • Symbolic repair (ASP) for coherence
+30%answer accuracy vs. vector RAG
<20msin-memory graph traversal
Multi-hopreasoning across sources
Real-timegraph updates, zero reindex

For technical teams

API-first, open, observable

Orpheam is not a black box. REST, MCP, webhooks, SDKs: everything is instrumented to live inside your stack, not next to it.

REST endpointsDocuments, graphs, answers, audit
MCP protocolPlug Cursor, Claude, ChatGPT
Signed webhooksGraph + answer events
ObservabilityOpenTelemetry · structured logs
curl -X POST https://api.orpheam.com/v1/answers \
  -H "Authorization: Bearer $ORPHEAM_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "workspace": "compliance-eu",
    "question": "DORA obligations for a critical provider?",
    "policies": ["compliance.read"],
    "citations": true
  }'

News & insights

From the blog

Production lessons, GraphRAG patterns and product updates — written for engineers and compliance teams.

View all articles
GraphRAG

Why vector RAG breaks the moment an auditor walks in

Vector chunking unlocks demos. It collapses on the three questions a regulator always asks: where is the proof, who could see it, and can you replay it?

June 4, 2026 · 7 min readRead
Compliance

Per-query ACL: enforcing permissions before generation

A manager, an employee and HR ask the same question and get the same engine — scoped to what each is allowed to see. Here is how access travels with the query.

May 21, 2026 · 6 min readRead
Engineering

Incremental graph updates with Graphiti on Neo4j

Patch a single regulation and the graph follows — no full reindex. A look at delta streaming and version-per-entity in the Orpheam pipeline.

May 9, 2026 · 9 min readRead

Get started with Orpheam

Increase your RAG accuracy with expert guidance

Bring your documents and your constraints. We prepare a tailored demo on your data and map a concrete path to production.

Book a demoTalk to an expert