NM Intelligence — For IBs · VCs · Law Firms

Your conversation stays
on your servers. Always.

The only AI analysis tool built for deal professionals where your client names, deal codes, and matter details never leave your infrastructure. Only a sanitized brief reaches any external model. And that brief gets answered by multiple frontier models simultaneously — synthesised into one decisive, domain-calibrated answer.

Investment Banks
Commercial diligence · sector research · IC prep
Venture Capital
Deal evaluation · underwriting · portfolio intelligence
Law Firms
Clause analysis · diligence · risk synthesis
Request a demo See how it works ↓
Architecture — how every query flows through NM Intelligence
🔒 On your servers Primary model · Sanitization · Session memory · Deal intelligence ↗ Sanitized brief only
Query
Your analyst
IB · VC · Law
Your servers · Always
Primary model
Full conversation · Memory
Your servers · Strip
Sanitize locally
Entities removed · Gate
Frontier model 1
Your choice
Frontier model 2
Your choice
Synthesise
Synthesis model
Adjudicate · Compress
Final answer
High-signal
Better than either alone
Works with GPT · Claude · Gemini · Harvey · or any frontier model your firm has approved

🔒
Your entire conversation stays on your servers. Always.
The primary model, session history, deal memory, and context — all handled on your own infrastructure. Only a sanitized analytical brief — with entity names already stripped — reaches any external model. Every query is logged with a complete diff of what was sent vs what was masked. No other AI tool for deal professionals can say this.
Conversation on your servers
Deal memory on your servers
Only sanitized brief sent externally
Full audit trail — original vs sanitized diff
Exportable for compliance review

The problem

Single models give single opinions

Your analysts use whatever AI tool is available — and switch to personal ChatGPT when it isn't good enough. Every existing tool gives one model's view. None of that is good enough for decisions with real stakes.

📱

Associates use personal ChatGPT

Because their firm's AI tool isn't good enough. Client names, matter details, and deal terms flow out on personal accounts with no audit trail and no control.

🎲

One model, one perspective

Every legal AI tool today gives you one model's answer. That model hedges, misses angles, and can be confidently wrong. There is no verification layer.

🔓

Confidential data leaves the room

Client names, deal codes, counterparty terms, and internal analysis go to external servers with every query. Your compliance team has no visibility.

🧠

Deal intelligence disappears

Five people working a deal for six weeks. When it's done, all the analysis, risks flagged, and decisions made — gone. The next deal starts from zero.


How it works

From query to synthesis in four steps

NM Intelligence adds a governed multi-model reasoning layer on top of NeutralModel's existing access controls, masking, and audit infrastructure.

01
🔒

Everything starts on your servers

Your entire conversation — context, follow-ups, session history — is handled by a model running on your own servers. Nothing about your interaction reaches any external service. Only a sanitized analytical brief ever leaves your perimeter.

02
✂️

Sanitize before anything leaves

Before the analytical brief reaches any frontier model, a local model strips company names, deal codes, people, and exact figures. The external model sees the substance of your question — never its identity.

03

Parallel frontier model calls

The sanitized brief goes to two frontier models simultaneously — not sequentially. Works with any models your firm has approved. GPT, Claude, Gemini, Harvey — you choose. Both answer in parallel, cutting latency in half.

04
🧠

Synthesis adjudicates

A synthesis model compares both answers against your original query. It discards weak points, resolves disagreements, and returns one answer stronger than either model alone — back on your servers.

05
📋

Full audit trail

Every turn logged on your servers — original query, sanitized brief, both model answers, and synthesis. Your compliance team sees exactly what was sent where and what came back. Exportable on demand.


Why synthesis wins

Better answers. Calibrated for your domain.

Individual models hedge and give generic answers. NM Intelligence synthesises two frontier models and calibrates the output for your specific domain — IB, VC, or law. The answer an IB analyst gets for a diligence question looks like what a senior colleague would put in an IC memo. Not a generic AI response.

Sample query — Investment Banking domain
"What are the top diligence risks for a SaaS company with 41% revenue concentration in one customer?"
Model A alone
6.7
Model B alone
6.8
NM Synthesis
9.0
Scored on decisiveness, actionability, insight depth, and compression. Synthesis added an original insight not present in either model — the 70% ARR threshold for covenant triggers. Structured as an IB analyst would write it, not as a chatbot would explain it.
🎯

Decisive, not hedged

Individual models give you both hands. NM Intelligence gives you a view. Synthesis commits to a judgment — ranked by materiality, structured for a decision. That's what deal work requires.

🏛️

Domain-calibrated output

NM Intelligence knows the difference between an IC memo and a reference call prep. The answer structure, language, and framing changes based on whether you're in IB, VC, or law — not one-size-fits-all AI output.

💡

Adds original insight

Synthesis consistently surfaces insights not present in either individual model — derived from the intersection of both answers and the original query context.


Use cases

Built for how deal professionals actually work

NM Intelligence is purpose-built for the three segments where analytical quality and confidentiality are non-negotiable at once.

Deal intelligence that stays on your servers

Associates and VPs spend hours on commercial diligence, sector research, and IC prep. NM Intelligence lets them query multiple frontier models at full capability — with answers structured the way an IB analyst actually needs them. Ranked by materiality. Framed for a decision. Not a generic AI response.

Deal names, target companies, and client identities never leave your servers. Five people can work on the same deal from one shared deal room — context accumulates, no one starts from zero.

  • Answers structured for IB — IC memo style, not chatbot style
  • Deal rooms — shared folder, five people, one deal context
  • Deal names, codes, and counterparty names stripped on your servers
  • Parallel frontier model calls — your choice of models
  • Every query logged with original vs sanitized diff
  • Works inside Microsoft Word — no context switching
09:13SERVERFull query handled on your servers
09:14SANITIZE"Project Falcon" → "the target company"
09:14SANITIZEStrategic investor → "lead investor"
09:14SANITIZE₹1,480 crore → "high valuation"
09:15MODEL 1Frontier model → answer received
09:15MODEL 2Frontier model → answer received
09:15SYNTHIB synthesis → IC memo structure · high signal
09:15AUDITFull diff logged · user: rahul.mehta@axisbank.com

Portfolio intelligence that compounds across sessions

NM Intelligence builds persistent memory of every company your team analyses — and answers are structured the way a VC associate actually writes. Investment thesis framing, downside risk weighting, moat assessment. Not a generic analysis.

Five people can work a deal from one shared deal room. Context accumulates across sessions — risks flagged on Tuesday surface automatically on Friday. Deal memory stays on your own servers.

  • Answers structured for VC — thesis, moat, downside, not generic output
  • Deal rooms — shared team folder, shared context, one deal
  • Cross-session memory per company — no one starts from zero
  • Confidential routing for portfolio company names and fund terms
  • All deal intelligence on your own servers
MonMEMORYHelix Bio session → 3 risks extracted
TueMEMORYMgmt concern flagged → stored to deal profile
FriMEMORYNew session → prior context injected
FriSANITIZE"Helix Biologics" → "biotech target"
FriMODEL 1Frontier model → parallel answers
FriMODEL 2Frontier model → parallel answers
FriSYNTHVC synthesis → grounded in 4 prior sessions

Privileged matters stay privileged. Always.

Attorney-client privilege demands confidentiality. NM Intelligence enforces it at the infrastructure layer — not as a policy, as an architecture. Your entire conversation stays on your servers. Client names, matter references, and case codes are stripped before any external call.

Answers are structured for legal work — issue hierarchy first, exposure ranked by severity, practical next steps. Not a generic AI answer reformatted. Works directly inside Microsoft Word — highlight a clause, synthesis appears in the side panel.

  • Answers structured for law — issue hierarchy, exposure, next steps
  • Deal rooms — matter-scoped shared folder for the whole team
  • Microsoft Word add-in — synthesis inside your document
  • Client names and matter codes stripped at source
  • Full audit log — complete record for regulatory and internal review
10:02SANITIZE"IndoBridge MedTech" → "the distributor"
10:02SANITIZE"Matter #2024-881" → stripped
10:03MODEL 1Frontier model → parallel answers
10:03MODEL 2Frontier model → parallel answers
10:03SYNTHLegal synthesis → issue hierarchy ranked
10:03PRIVThread: private · not shared with team
10:03AUDITFull record logged · user: priya.iyer@cam.co.in

Microsoft Word

NM Intelligence lives inside Word

Your lawyers and analysts work in Word. NM Intelligence works with them — directly in the document, no context switching, no browser tab, no phone.

✍️

Highlight any text

Select a clause, a paragraph, or the entire document. NM Intelligence reads what you selected and sends it through the synthesis pipeline.

🔒

Client names never leave your servers

The local sanitization model strips identifying information on your servers before anything reaches any external model. Your client's name, matter reference, and deal code stay where they belong.

Zero friction — already authenticated

No login prompt. The add-in recognises who you are from your corporate identity. Open Word, the panel is there, start working.

WordClause 9.4 selected · 340 chars
SANITIZE"Project Falcon" → "the transaction"
SANITIZE₹620 crore → "deal value"
EXPERTFrontier models → answering in parallel
SYNTHSynthesis ready in task pane

OutputCORE TAKEAWAY:
Broad indemnity with no cap creates open-ended seller liability. Renegotiate before proceeding.

AuditLOGuser: priya.iyer@cam.co.in · matter: redacted

Deal intelligence
Knows what's confidential.
Before your analyst types a word.

Until now, firms working on sensitive deals were stuck with weaker local models — because data could not leave. NM Intelligence ends that trade-off. When a deal room is created, it automatically scrapes the company's public website and builds a keyword profile. Every query is checked against it — deal-sensitive queries go through the sanitization layer first, everything else reaches frontier models directly.

Your team gets full frontier model quality on confidential work. Your admin stays in control — they can review, add, or remove keywords at any time.

🌐
Auto-seeded at room creation. Enter the deal company when creating a room. NM scrapes their public website, extracts entities and terms, and arms the sanitization layer — before the first analyst opens the room.
🔀
Intelligent per-query routing. Query matches a deal entity — sanitization triggers, names are rewritten, a clean brief reaches frontier models. No match — direct frontier model call, full quality, no friction.
🔧
Admin stays in control. Review the keyword list, add anything missed, remove anything irrelevant. The system enforces it for every team member, every query, automatically.
09:01CREATEDeal room opened · company website entered
09:01SCRAPEPublic site → 43 entities extracted
09:01ARMEDKeyword profile active · reprompt layer ready
09:14MATCHDeal entity in query → sanitization triggered
09:14REWRITE"Target Co" → "the target company"
09:22NO MATCHGeneral query → direct frontier model call

Deal rooms

Five people. One deal. Shared context.

Every deal has a shared folder. Associates, VPs, and partners all work from the same space. Context accumulates across sessions — no one starts from zero when they open a new chat on a live matter.

📁

Shared deal folder

Create a deal room, add your team. Every session inside the room is scoped to that deal. Context, flags, and analysis stay inside the folder — not scattered across individual chat histories.

🧠

Memory that compounds

Risks flagged on Monday surface automatically on Friday. Entity profiles build over time — the longer the team works a deal, the more context NM Intelligence has to work with. All on your servers.

🔒

Private threads within the room

Associates can work privately inside a deal room. Their individual analysis stays theirs until they choose to contribute it. Partners see the shared deal context — not every individual session.

📋

Full audit per deal

Every query, every sanitization, every synthesis — scoped to the deal and attributable to the team member. When compliance asks what AI was used on a matter, you have the complete record.


Deployment

Your infrastructure. Your data. Your container.

NM Intelligence runs as a self-hosted Docker deployment on your own servers. The primary model, sanitization layer, synthesis engine, and all deal memory run on infrastructure you own and control. NeutralModel disappears — the system runs without us.

Sanitization runs on your servers

The local model strips entity names and confidential terms before any external API call. Deal names never reach any frontier model — only a sanitized analytical brief leaves your servers.

Inspect the container before you sign

Download the Docker image, inspect every file, run it on your own servers or cloud account. We encourage your security team to review it end-to-end before signing anything.

Deal memory stays on your servers

Every entity profile, session summary, and extracted insight is stored inside your own infrastructure. NeutralModel has zero access to your deal intelligence.

If NeutralModel disappears, you still run

The container is self-contained. Your team continues working even if NeutralModel ceases to exist. No operational dependency on our uptime.

$docker pull neutralmodel/intelligence
Pulling from neutralmodel/intelligence
Status: Image is up to date

$docker run -d \
--name nm-intelligence \
--gpus all \
-p 8080:8080 \
-v ./config:/config \
-v ./deal-memory:/data \
neutralmodel/intelligence

NM Intelligence running on :8080
Primary model loaded · your servers
Deal memory initialised · your data, your servers
Air-gap ready · No NM dependency

Standard vs Intelligence

Which tier is right for your organisation?

NM Standard governs AI access across your organisation. NM Intelligence adds a multi-model synthesis layer for teams where analytical quality and confidentiality are both non-negotiable.

Capability Direct to LLM NM Standard NM Intelligence
PII masking
Company keyword masking
Full audit logs
Per-user cost control
Docker / VPC deployment
Multi-model synthesis (any frontier models)
Deal-aware keyword routing
Auto-seeded entity profiles at room creation
Parallel expert calls
Cross-session deal memory
Microsoft Word add-in
Original vs sanitized query diff log
Best for Ad hoc use Enterprise governance IB · VC · Law

Early access — select firms only

See NM Intelligence on your deals

Book a 30-minute walkthrough tailored to your firm's deal workflow, governance requirements, and infrastructure. No commitment required.

hello@neutralmodel.com · We respond within one business day