nBrain builds one private AI Brain around your ledgers, deal files, client records, and policies. It runs in your own cloud account or on your own servers, respects the information barriers between deals and clients, keeps the records your regulators expect, and waits for a named approver before anything is paid, sent, or filed.
The same platform is already running real work for nBrain clients.
Results from Brains in production for nBrain clients across industries, not only financial services. Other client names are withheld.
Written for the CFO, CCO, or CISO who has to sign off. nBrain supports your compliance program. It doesn't replace it, and it runs inside the controls you already have instead of adding a new place for client data to live.
Most firms run the Brain in their own cloud account. Firms with the most sensitive deal or client data put it on their own servers, or on a network with no internet at all. The Brain, its controls, and its audit trail are the same in all three.
The Brain runs in a region you choose, under your accounts, your keys, and your billing. You see every resource and every model call.
Installed on hardware you own, behind your firewall. Sensitive work stays on local models, and cloud models are used only for work your policy allows.
A network with no route to the internet, for deal data or client files that must never leave the building.
GPT-6 Astra, Claude Opus 5.5, and Gemini 4 Argon are hosted services, so a disconnected Brain uses open-weight models on your own hardware instead. Those models handle search, cited answers, summaries, extraction, and first drafts well. On the hardest multi-step analysis they can trail the frontier models, so before go-live we test them on your own documents and show you where the difference matters.
Access set per deal, fund, client, and entity, from your identity provider. A deal team never sees another team's file.
Personal and payment identifiers are masked before indexing, or kept to models running inside your environment.
Payments, client communications, filings, and record changes wait for a named approver. Two where your policy calls for two.
Every prompt, source, model, user, and action, logged in your environment and kept on your retention schedule.
How financial firms can use frontier AI when part of their data can't leave the building: where it runs, who can see what, how records are kept, and which workflows to start with.
Each team starts with one costly, everyday job. The Brain does the reading, matching, checking, and first drafts on your own data. Your people make the calls, and approve anything that moves money or reaches a client.
The Brain pulls bills out of the shared AP mailbox, reads every line, matches it to the purchase order and the receipt, suggests the coding, and queues it for approval. Duplicates, price and quantity mismatches, missing POs, and changed bank details are flagged before a payment is ever prepared.
The Brain reads your accounting system, bank feeds, and spend tools for every entity, matches transactions each day, and puts the unmatched items in front of your accountant with the likely cause. It drafts the variance commentary and the cash forecast, and your controller signs off.
For credit teams, the Brain reviews each order on credit hold against the signals your analysts weigh by hand: tolerance by customer tier, unapplied cash, pending credits, and how the customer actually pays. For lenders, it checks each loan file against your own product rules and lists what's missing before an underwriter spends an hour on it.
The Brain reads data rooms, fund documents, side letters, and past investment memos, answers the deal team's questions with the page cited, and drafts first-pass diligence summaries, DDQ responses, and quarterly LP letters. Each deal sits behind its own wall, so only the people on it can search it.
Before each meeting the Brain pulls together the client's accounts, last meeting notes, open tasks, life events, and planning documents into one brief. After the meeting it drafts the notes and follow-up for the advisor to approve, and keeps both as part of the client's record.
The Brain reads broker submissions, loss runs, applications, and inspection reports, checks them against your appetite and guidelines, and hands the underwriter a summary with what's missing. On claims, it builds the file summary from notes, photos, and documents, and flags coverage questions for the adjuster.
The Brain reads new orders from agents and lenders, opens the file, builds the closing checklist, and chases missing documents. It reads every email that mentions wiring funds, flags changed instructions and look-alike senders, and holds disbursement until a person has verified by phone.
Staff ask the Brain about your policies, procedures, and regulatory guidance and get the answer with the exact section. For audits and exams, it gathers the evidence, maps it to each request, and drafts the response for your compliance team to review.
Your systems feed one knowledge layer. The Brain routes each task to the right model and takes action in the tools your staff already use, under the rules you set.
Stood up in your own cloud account, on your own servers, or on a network with no internet.
Your accounting, banking, CRM, and document systems, read-only by default.
Agents read, match, check, and draft. A person approves anything that moves money or reaches a client.
We take a baseline in month one, then add the next workflow on the same data and permissions.
The first question a CFO or CCO asks isn't "what can it do?" It's "where does client data go, who can see it, and what happens when it's wrong?" These controls are built in from the first week.
In your cloud account, on-prem, or air-gapped, so client and deal data stay where you already protect them. Your data never trains public models.
Walls between deals, funds, clients, and entities, mirrored from your identity provider and checked on every request.
Payments, wires, client communications, and record changes wait in a queue until the right person, or two, approves.
Which data, which model, which person, and which action, recorded every time, ready for your examiner, auditor, or LP.
Prompts, outputs, and logs follow your retention schedules and legal holds, and can be produced on request.
What we do and don't claim. nBrain doesn't hold a certification or attestation that covers your data. The Brain deploys inside an environment you already control and audit, so it sits within your existing controls like any other system you run. Your data stays in your environment, and our staff reach it only through accounts you issue and can revoke.
Each agent has one job, a defined set of tools, and guardrails on what it can do. They share the same data, walls, and audit trail, so each new one ships faster than the last.
Reads each bill, matches it to the PO and receipt, and suggests the coding.
Flags duplicates, look-alike senders, and new bank details, and holds them for a call-back.
Matches bank, card, and payroll activity to the books every day, by entity.
Drafts variance commentary and cash forecasts from the ledger.
Scores each hold on your signals and recommends, with the reasons shown.
Cited answers from data rooms, and DDQs drafted from approved responses.
Drafts quarterly letters and client reviews from portfolio and account data.
Checks outgoing letters and posts against your rules, and answers policy questions.
Every agent is built for your environment and your rules, not pulled from a shared library. What is a Centralized AI Brain? →
These come from Brains nBrain has built or is building for clients today. Each one maps to work finance, credit, and investment teams do every day.
A Tier 2 defense manufacturer's AP clerk used to work the invoice mailbox by hand. Now each bill is pulled out, matched to its PO and receipt, checked for duplicates and amount mismatches, and queued for approval.
The Brain runs on a server in the firm's own office, so sensitive deal and banking mail never leaves the building. It sorts 19 years of email, and quarantines phishing and wire-fraud lures for review instead of filing them. Daily bank-against-books reconciliation is in build.
The billing team's Brain checks each field ticket against payroll time, GPS time, and the required safety paperwork, sends mismatches to an exception queue, and assembles one invoice package per job. Each company's staff see only their own company's data.
Results are anonymized except where the client has agreed to be named. Figures are from client work and are labeled where they come from evaluation or are still in build.
Not to their public services, and never to train their models. Where you allow frontier models, the Brain calls private endpoints in your own cloud account, such as Azure OpenAI or AWS Bedrock, with zero retention where the provider offers it. For anything you don't want leaving your hardware, it uses open-weight models running in your environment. You decide which models may see which data, and every call is logged.
It is built to support them. Every prompt, source, output, and approval is logged in your environment, your retention schedules and legal holds apply, and anything that becomes a client communication can be sent to the archive you already use for email and chat. Nothing reaches a client without a person approving it. Your compliance team decides what counts as a record and how long it is kept.
nBrain doesn't hold a SOC 2 report of its own today. For most firms that isn't the question that matters, because nBrain doesn't host your Brain. It runs inside your environment, so it falls within the controls your SOC 2 auditor already tests. We provide the architecture, data-flow diagrams, and control descriptions your auditor and your LPs' DDQs will ask for.
The Brain runs on infrastructure you already protect, so it inherits your encryption, MFA, access controls, and monitoring. Access follows your identity provider, nonpublic personal information can be masked or kept to models inside your environment, and logs stream to your SIEM for incident response. We document it for your written security program and your vendor-oversight file.
Yes. We configure walls per deal, fund, client, or entity, and every request is checked against them, including summaries that draw on many documents. Permissions come from your identity provider and source systems, so people only see what they can already see. Crossing a wall requires compliance approval, and it is logged.
No. Agents read, match, check, and prepare. They can't release a payment or a wire. Every payment waits for the approvers your policy requires, two where it calls for two, and every approval is recorded.
It checks every invoice and payment request before a person approves it: duplicates, amounts that don't match the PO, look-alike sender addresses, and new or changed bank instructions. We build it so a bank-detail change opens a call-back task to a number already on file. Suspicious messages are quarantined for review instead of being filed. It reduces risk. It doesn't replace your verification procedures.
The Brain is designed to stay out of your cardholder data environment. Card numbers can be detected and masked before anything is indexed or sent to a model, and payment systems connect read-only to tokenized fields. If a workflow genuinely needs card data, we scope it with your PCI assessor first.
Yes. You choose the region, and the Brain, its indexes, logs, backups, and model endpoints stay in it. For the most sensitive data it can run on your own servers, or on a network with no internet connection.
It will sometimes be wrong, so it's built for that. Answers show their sources, low-confidence answers are flagged and sent to a person, and anything that moves money or reaches a client waits for approval. Every step is logged, so you can see what happened and why, and corrections become rules the Brain follows next time.
Builds start at $40,000, and a typical build takes about three months. Running costs, meaning your own cloud and model accounts, are billed to you directly at list price, with no markup. Because each task is routed to the right model, model usage typically costs about 65% less than sending everything to one top-tier model.
With one high-volume workflow where people spend their day reading, matching, and re-keying: AP invoice matching, bank reconciliation, credit holds, loan file checks, DDQs, or client review prep. We take a baseline in month one and measure hours returned, then add the next workflow on the same foundation.
Your team can build a chatbot. Most of the work is in what sits around it: permission-aware search, information barriers, model routing, evaluation, approval gates, audit logs, and keeping all of it current as models change. We bring that layer already built. Your IT team stays involved, owns the code and data, and can take it over at any time.
How to put frontier AI to work when part of your data can't leave the building.
Read the guide →
Finance, the factory floor, and engineering on one Brain, starting with the AP mailbox.
Read the story →
Where the Brain runs, who can see what, and how every action is recorded.
Read the overview →
32,000+ client interactions and the operating model behind every successful build.
Read the report →In one strategy session we map where a Brain would give your team the most time back, which deployment fits your data, and what your CCO, CISO, auditor, and LPs will want to see.
Prefer email? hello@nbrain.ai