MIRNETHresearch intelligence

Est. 2021 · London & Boston · SOC 2 Type II

Answers your auditors
will accept.

Mirneth is a research assistant for teams whose work cannot cite a hallucination. Every sentence it returns is traceable to a passage in your own corpus — with the paragraph, the page and the version.

Deployment
Cloud, VPC or air-gapped
Corpus size
Up to 40M documents
Citation recall
99.2% verified
RESEARCH MEMO 04-118 Draft

“What is our exposure if the EU AI Act classifies this feature as high-risk?”

1. Under Article 6(1), a system is high-risk where it is a safety component of a regulated product.[1] Our classification module is not a safety component; it informs human decisions.[2]

2. Obligations that still apply include documentation (Art. 11)[3] and logging (Art. 12).[4]

3. Where the feature is later extended to biometric identification, the analysis changes materially…

Sources (4)
1Regulation (EU) 2024/1689, Art. 6 — p.31verified
2Internal memo, Legal/2026-14 — §2.1verified
3Regulation (EU) 2024/1689, Art. 11 — p.44verified
4Regulation (EU) 2024/1689, Art. 12 — p.46verified

In use at institutions including

Harrow & Voss LLPNorthgate HealthVercept CapitalKingsmere UniversityAtlas-9 Research
A researcher reading a document at a long table in a university reading room with tall arched windows
Fig. 1 Reading room, Kingsmere University. Corpora are indexed in place — cloud, VPC or air-gapped — and nothing leaves the customer's perimeter.
01

Method

Retrieval first, generation second. The model may only write sentences it can support.

Grounded generation

Every claim is mapped to a retrieved span before it is written. Unsupported sentences are deleted, not softened.

Deterministic retrieval

Pinned embedding model and index version per workspace. The same query on the same corpus returns the same sources.

Human-readable audit

Export a full reasoning trail — query, retrieved passages, rejected passages and final output — as a signed PDF.

Refusal by design

When confidence falls below the configured threshold, Mirneth states what is missing instead of improvising an answer.

02

Evidence

Third-party evaluation on a 12,000-question benchmark drawn from legal, clinical and financial corpora.

Table 1 — Benchmark results, Q3 2026 (n = 12,000)
SystemCitation recallUnsupported claimsRefusal accuracy
Mirneth 499.2%0.4%97.6%
General assistant A71.4%9.8%62.1%
General assistant B64.0%12.6%58.4%
Keyword + LLM baseline81.3%6.1%70.2%

Methodology and raw data available under NDA. Benchmark maintained by the Mirneth research group; auditors are independent.

Stacked archival books, an open bound report, reading glasses and a fountain pen on a wooden table
Fig. 2 Evidence chain — every returned sentence carries a document hash, page and character range that an auditor can re-open.
03

Workflow

A research task, from intake to signed memo.

A legal researcher standing in a scholarly library of bound volumes

“The memo arrives with its evidence attached. My partners stopped asking me to re-check the citations.”

Dr. Adrian Vale · Head of Knowledge, Harrow & Voss LLP

  1. i

    Ingest

    Connect a document store, DMS or data room. OCR, layout parsing and redaction run on ingest; nothing is retrained.

  2. ii

    Scope

    Select the corpus subset and the matter or project. Permissions are inherited from the source system, not re-declared.

  3. iii

    Interrogate

    Ask in natural language or build a structured query. Each answer shows its sources inline with page-level anchors.

  4. iv

    Certify

    Export the memo with a provenance appendix and a hash of the corpus version used. Reviewers sign inside Mirneth.

04

Security

Built for buyers who read the DPA before the demo.

Data residency

EU (Frankfurt), UK (London), US (Virginia) or your own VPC.

Access control

SAML SSO, SCIM provisioning, and per-document permissions mirrored from source.

Encryption

AES-256 at rest with per-tenant keys; TLS 1.3 in transit. BYOK available.

No training

Your corpus and queries are never used to train or fine-tune any model.

Retention

Configurable from 24 hours to 7 years, with legal hold support.

Certifications

SOC 2 Type II, ISO 27001, ISO 42001 (AI management). HIPAA and GDPR ready.

05

Licensing

Annual agreements. Seat bands, not per-question meters.

Practice

Small teams, single corpus

£1,200/mo

  • 10 seats
  • 500k documents
  • Cloud EU or UK
  • Standard support
Request access

Enterprise

Regulated, multi-jurisdiction

Bespoke

  • Unlimited seats
  • 40M documents
  • Air-gapped option
  • Custom evaluation suite
  • Audit support included
Contact us
06

Notes

1 What counts as a “verifiable” citation?

A citation is verifiable when the quoted span exists in the indexed version of the source and semantically supports the claim in the sentence. Each citation carries a document hash, page and character range.

2 Can we bring our own model?

Yes. Mirneth supports model routing to your own provider endpoints, including on-premise deployments, while keeping the retrieval and verification layer intact.

3 How long does implementation take?

Typical cloud deployments reach first value in ten working days; VPC and air-gapped deployments run six to nine weeks including security review.

4 Do you support non-English corpora?

The retrieval layer handles 24 languages and cross-lingual queries; generated memos are produced in the language of the requesting user.

Bring a hard question to the briefing.

We will run it live against a sample corpus in your industry and show you every source behind every sentence.