Doxwell

For platform partners

The document brain for ECM, IDP, and VDR platforms

Embed Doxwell's engine in your product. MCP-native, air-gap-capable, audit-grade. Built on the same engine that runs Doxwell.

Talk to Pavel about embedding Doxwell

What it is

Docortex is the engine inside Doxwell. We expose it as a licensable component: same code, your infrastructure.

It runs air-gap. It speaks MCP natively. It produces citations, not vibes.

What it does that vanilla RAG cannot

Temporal facts

Every fact carries a valid-from and valid-to. Ask "what was true at signing?" and get the right answer, not the latest.

Entity resolution

"Condor Immobilien GmbH", "Condor", and "C.I.P." are the same entity. Doxwell knows.

Contradiction detection

When document A says "closing 2024-06-30" and document B says "2024-09-15", Doxwell flags both — with provenance.

Audit-grade trace via RIGOR

Answers cite the source paragraph at query time. Independently, rigor verify re-derives the deterministic parts of the answer from disk alone — no second LLM call to audit.

By the numbers (today's nightly)

Documents
21,702
Entities
38,066
Facts
102,417
MCP tools
618
RIGOR invariants
22

Updated May 19, 2026. See /methodology for how we measure.

Where Doxwell fits in your stack

vs. vanilla RAG

We're the layer above retrieval. We don't replace your vector DB; we add the graph, the timeline, the provenance.

vs. Microsoft GraphRAG

GraphRAG (April 2024+) builds an entity-relationship graph at index time. We agree with the shape. Three things we built differently: (a) temporal validity per fact, queryable with valid-at <date>; (b) supersession events recorded as first-class objects (no destructive update); (c) citation-bound answers reconstructable from the ledger without re-calling the LLM. Our framing of the retrieval quality delta is +36% NDCG@10 over BM25 on our 21,702-doc nightly harness; we have not published a head-to-head benchmark, and replication on a shared open corpus is a roadmap item. See /methodology for our harness, corpus, and how we publish.

Microsoft GraphRAG on GitHub →

vs. LangChain / LlamaIndex

Frameworks for building. We're a deployable engine.

How partners integrate

MCP server

Drop-in for any MCP-compatible agent. ~30 minutes to first answer.

Embedded library

Python and TypeScript bindings. For partners building on top.

Co-branded surface

Your UI, our engine. White-label available.

Built for traceability

  • GDPR-aligned by design — local-first execution; processor agreement available.
  • Replayable RIGOR ledger — every answer reconstructable from disk, no LLM round-trip needed for audit.
  • Citation-bound answers — every claim resolves to a source paragraph in the database, not the prompt.
  • Signed binaries — minisign verification on every release.
  • On-prem deployable; air-gap supported. Cloud LLM is opt-in, not the default.

Embed it.

Tell us what you build and what you would put Doxwell inside. We reply within 2 business days.

Mention: company, role, what you would embed it in, expected document volume.

Talk to Pavel