Straight answers to the questions owners ask us first.

nBrain designs, builds, and operates private AI platforms for mid-market and regulated companies. Each one is a Centralized AI Brain: one platform connected to your data, your systems, and the way your team works, running in your own cloud or on your own servers.

You own it. We build it with you, train your team on it, and stay to run and improve it for as long as you want us to.

One AI platform for the whole company, instead of a separate AI tool in every department. It has four layers:

• Knowledge: your documents, email, calls, CRM, and operational data, so every answer is grounded in your business and cited back to the source.
• Model router: every frontier model behind one door, with each task sent to the model best suited to it.
• Memory: every correction and decision is kept, so you only teach it something once.
• Connections and audit: direct links into the systems you already run, with permissions checked and every action logged.

Your team uses it in a chat window, inside Slack, Teams, and email, and through agents that handle specific jobs. Read the full explainer.

No. The Brain is the platform. Agents are built inside it. An agent is a purpose-built worker with a defined role, the tools it may use, and guardrails, such as an RFQ agent or an accounts-payable check.

Every agent shares the same knowledge, memory, and permissions. That is why the second agent ships faster than the first, and the third faster again.

ChatGPT and Claude are excellent general models. Ask them about your business, though, and they are guessing. They don't know your customers, pricing, or history. They forget yesterday's context. And whatever your people paste in leaves the building.

The Brain uses the same frontier models, including GPT-6 Astra, Claude Opus 5.5, and Gemini 4 Argon. The difference is that it grounds every answer in your own data, remembers what it is taught, runs inside your boundary under your rules, and belongs to you.

Copilot is a strong assistant for Microsoft 365, priced per seat, and it mostly sees Microsoft's apps. A Brain connects every system you run, including your ERP, CRM, shop floor systems, and call recordings. It routes each task to the best model, takes approved actions, and belongs to you.

Many companies keep Copilot for writing in Word and Outlook. The Brain is the layer that knows the whole business.

Enterprise search products like Glean are strong at finding information across many apps, and they suit large enterprises that want one search layer across thousands of seats. They are sold per seat.

nBrain is built differently. We build a platform around one company's workflows rather than a horizontal search layer. You own the code, data, and infrastructure. There is no seat count, so a 150-person company isn't paying for seats it doesn't use. And we do the build and the adoption work with you, rather than leaving rollout to your team.

Each SaaS AI feature sees only its own app's data, bills separately, and follows its vendor's rules. Five tools means five partial views of your business and five invoices.

The Brain sits across all of them with one knowledge layer, one set of permissions, and one audit trail. It still uses those tools, through their APIs, where they are the right tool for the job.

Your team can build a chatbot. Building has never been easier. Where in-house efforts get stuck is the layer most teams don't know exists until they are in it: permission-aware search across systems, model routing, evaluation, approval gates, audit logs, memory that compounds instead of piling up noise, and keeping all of it current as models change every month.

We bring that layer already built and proven in production. Your IT team stays involved from the first week, owns the result, and can take it over whenever it wants.

Owner- and CEO-led mid-market companies, from a few dozen to a few hundred employees, often in regulated industries: defense manufacturing and government contracting, healthcare, financial services, construction and energy, and professional services.

It fits best when you have proprietary data, processes, or expertise that set you apart, and you want AI your company controls rather than a tool every competitor can buy.

If you need one off-the-shelf tool for one narrow task, a SaaS product will usually be cheaper and faster. If no one inside your company can own the program and give it a few hours a week, it is worth waiting until someone can. And very large enterprises that mainly want horizontal search across thousands of seats are often better served by enterprise search vendors.

You do. The code lives in your repository. The infrastructure runs in your cloud account or on your servers. The data, configurations, prompts, agent definitions, workflows, and integrations are yours, and so is the IP we build for you under your engagement.

You do, always. It stays in your environment, in systems you control, and it never trains anyone else's model.

You do. They are stored in your repository and your environment, written so your team can read and change them. Nothing that makes your Brain work is held back on our side.

Your Brain keeps running. It runs in your account, the code is in your repository, and the model accounts are in your name, so nothing depends on nBrain's servers. Your IT team, or another partner you choose, can operate and extend it.

Yes, at any time. We document the architecture, hand over every credential and runbook, and can train your IT team to run it. Your IT team is involved from the first week, so a handover is a continuation, not a rescue.

It is the question we hold every build to: if you ended the relationship tomorrow, would everything keep working, and could someone else run it? The answer has to be yes. That means code in your repository, infrastructure in your account, model contracts in your name, and documentation your team can follow.

No. There are no seat licenses, no proprietary hosting, and no single model vendor. Model and cloud usage runs in your own accounts. After the build, an ongoing engagement with us is your choice, not a requirement for the Brain to keep running.

Yes, and yes. Models are engines and they are swappable. Your data, memory, permissions, and workflows live outside the model, so they stay put when the model changes.

We have done it under pressure: when a model provider changed its policy in the middle of a live engagement, we moved the deployment to an open-weight model inside a day, with no rebuild.

Yes. Your team can edit instructions, templates, approval rules, and who sees what without waiting on us. Larger new workflows we build with you, or your IT team can build them on the same foundation, because the code is yours.

You do, workflow by workflow. Each one has a setting: suggest, draft, act with approval, or act and report. Payments, customer messages, publishing, and changes to your systems wait for a named person unless you decide otherwise, and every action is logged either way.

Only the parts of a task you allow, and only through enterprise endpoints that do not train on your data. For anything you don't want leaving your own hardware, the Brain uses open-weight models running inside your environment. You decide which models may see which data, and every call is logged.

No. Your data never trains public models. Model calls go through enterprise endpoints that don't train on what you send, or to models hosted entirely inside your environment.

Wherever your data is allowed to live: your own AWS, Azure, or Google Cloud tenant, a government cloud such as AWS GovCloud or Azure Government, servers in your own data center, or a fully air-gapped network. It runs in your account, not ours.

Yes. In an air-gapped deployment the Brain runs on your own servers and GPUs, with open-weight models such as Llama, Mistral, or Qwen running locally. Updates arrive through a controlled path your team approves.

The honest trade-off: GPT-6 Astra, Claude Opus 5.5, and Gemini 4 Argon are hosted services and can't run disconnected. Open-weight models handle search, cited answers, summaries, and drafting well, and can trail on the hardest reasoning, so we test on your documents before go-live and show you where it matters.

Access mirrors your company sign-in and the permissions in each source system. If someone can't open a file today, the Brain can't open it for them either. Financial, HR, and owner-level information can be restricted to the people who should see it while everyone else uses the same Brain. Permission changes are logged.

Yes. Every question, the sources used, the model that answered, the person who asked, and any action taken are recorded in your environment. You can answer an auditor or your board with a record, not a guess.

The Brain treats emails and documents as information, never as instructions. Anything that sends, pays, submits, or changes a system is locked by our servers, not left to the AI's judgment, and waits in an approval queue for a person. Unknown or made-up tools are refused outright. The worst case is a wrong answer on screen, never an action behind your back. See how security works.

No, and we don't claim to be. For most deployments we don't need to be, because nBrain doesn't run your Brain on our servers. It deploys inside an environment you already control and have already assessed, so it falls within your existing controls. We provide the architecture, data-flow diagrams, and control documentation your auditor or assessor will ask for.

It is built for ITAR and CMMC environments. There is no ITAR certification for any product to hold; ITAR is a regulation your company complies with, and the Brain is designed to support that compliance. It deploys inside your existing boundary, on US infrastructure you control, and access to controlled technical data follows the authorizations your system already maintains. See private AI for defense and government.

HIPAA is a law your organization complies with, and the Brain is designed to support that: it runs in your environment, follows role-based access and the minimum-necessary principle, encrypts data in transit and at rest, and logs every access. We sign a BAA where PHI is involved. See private AI for healthcare.

Only the people working on your build, only through accounts you issue, and only for the scope and time you grant. You can revoke that access at any time, and every session is recorded in your audit trail.

Security is a routine we run, not a one-time setup. Critical weaknesses are patched first, as soon as a fix is tested, and high-risk issues follow on a set schedule. We are rolling out attack testing against known trick emails and documents before every update, weekly reviews of new security alerts, automatic scanning for weak software and exposed keys, and a regular break-in test with a written report to you.

It will sometimes be wrong, the same way a new hire is, so it is built for that. Answers show their sources, low-confidence answers are flagged, and anything that matters, such as a payment, a customer message, or a published page, waits for a person to approve it. When you correct it, the correction is kept. Every step is logged, so you can see what happened and why.

Builds start at $40,000. We scope the build with you before anything starts, so you know the price before you commit. Where it lands above the starting price depends on how many systems we connect and how many workflows go live.

A typical build takes about 3 months from kickoff to a Brain your team uses every day. The first working version on your real data comes well before that, so your team is testing it in the first weeks rather than waiting for a launch day.

The foundation: the Brain stood up in your environment, your first data sources connected, your first workflows or agents live on real work, permissions, approval gates and an audit trail, and training for each role that uses it. We confirm the exact scope in writing before the build starts.

Mostly scope: the number of systems we connect, older systems without APIs that need custom connectors, the number of workflows going live in the first build, on-premises or air-gapped hardware, and the compliance documentation your environment requires. We would rather start with one or two workflows and add more on the same foundation than build everything at once.

Two kinds. First, running costs: your own cloud and model accounts, billed directly to you at list price, with no markup from us. Second, optional ongoing work where we operate and evolve the Brain (monitoring, tuning, model upgrades, and new workflows), scoped and priced to your needs. The Brain keeps running without it.

It depends on how many people use the Brain and for what. We estimate it with you during scoping. Because it is billed in your own accounts, you see every dollar, and the model router keeps routine work on lower-cost models, which cuts the cost per AI task by about 65%.

No. There are no seat licenses. Adding people to the Brain doesn't add license fees; it only adds the model usage they generate, billed at list price in your own accounts.

Per-seat tools charge for every employee, every month, whether they use the tool or not, and the bill grows with headcount. A Brain has no per-seat licenses. The model router sends each task to the right model and cuts the cost per AI task by about 65%. One platform replaces overlapping tools in each department. And what you build stays with you as an asset instead of resetting when a contract ends.

The first build is the foundation. Later modules, such as a new department's workflows, are scoped separately once the foundation is running, because by then we both know far more about what it will take. Each one reuses everything already built.

A typical build runs like this:

• Days 1 to 30: we confirm the first workflows and a written definition of done, take a baseline, stand the Brain up in your environment, connect the first data sources, and get a first working version in front of your team on real data.
• Days 31 to 60: the first workflows go into daily use with a pilot group. We tune answers, approvals, and permissions from what they find, and train each role.
• Days 61 to 90: rollout to the wider team, results measured against the baseline, and the next workflow chosen.

Your proposal spells out the exact plan for your build.

Less than people expect, but it has to be the right people. You need a named owner who runs the program week to week, an executive who sets priorities on a fixed rhythm, and a few hours a week from the people who do the work, to test the Brain on real cases. In our study of 32,000+ client interactions, every successful engagement had those in place at kickoff. Read The Other 50%.

• A named owner on your side
• One or two workflows to start with, and a written definition of what done looks like
• Access to the systems in scope
• Testing time from the people who will use it

With one high-volume workflow that causes pain every day and has a group of people waiting for relief, such as RFQs and proposals, call recaps, invoice checks, or client reporting. Daily pain drives adoption. Convenience tools tend to go unused. We take a baseline in month one, measure the hours returned, then add the next workflow on the same foundation.

No. The Brain is built to read your business in the form it already exists: documents, email, call recordings, spreadsheets, and system records. What does help is stable systems. If you are replacing your ERP or CRM in the next few months, we plan the build around that change.

It will sometimes be wrong, which is why the work that matters waits for approval. Open-weight models in air-gapped deployments can trail frontier models on the hardest reasoning. Old systems without APIs take longer to connect. And the hardest part is rarely the code. Projects stall when no one owns them, when there is no written definition of done, when too much is built at once, or when no one has time to test. We will tell you up front if we see those risks.

We start with work people already find painful, put the Brain inside the tools they already use, and train each role on its own work, not on generic prompting. We stay through adoption, and we don't call anything live until your staff use it unaided.

Yes. Training is role by role, in working sessions on your team's real work: what a salesperson asks it is different from what a controller or plant manager asks it. We keep training as new workflows go live.

In a chat window, and inside Slack, Microsoft Teams, email, and your CRM. People can type, talk, or drop in a file. There is no new app they have to remember to open.

Most AI waits for a prompt. A proactive Brain notices what needs to happen next, prepares the work, and moves it forward once a person approves. A call ends, and the recap, CRM update, and follow-up are drafted. An invoice fails a check, and it is held for review. See proactive AI in action.

Only what you have told it it may do on its own, and only after you have seen it do that work well. Payments, customer messages, publishing, and changes to your systems wait for a named person unless you decide otherwise.

Two ways. New material flows in continuously, such as this morning's meeting, yesterday's signed contract, or the CRM change from an hour ago, so its picture of the business stays current. And every correction is kept: tell it once that a client's terms are non-standard, and everyone benefits from that from then on.

All of the leading ones: GPT-6 Astra, Claude Opus 5.5, Gemini 4 Argon, and open-source models such as Llama, Mistral, and Qwen for private or on-premises work, plus specialized models for transcription, embeddings, voice, and images. The Brain is model-agnostic, so it uses whichever model does each task best.

Each request is assessed for complexity, context length, accuracy requirement, privacy, and cost, then sent to the right model. Deep reasoning goes to one, long documents to another, image and voice work to a third, and sensitive or routine work to an open-weight model inside your walls. Routing is per request, not per subscription, and it cuts the cost per AI task by about 65%.

Yes, in both directions. PDFs, images, screenshots, audio, video, and spreadsheets go in. Documents, decks, tables, graphics, dashboards, and voice come out.

Anything with an API, a webhook, or a database: Microsoft 365, Google Workspace, Salesforce, HubSpot, NetSuite, SAP, QuickBooks, SharePoint, Slack, Teams, Zoom, Snowflake, and more. Older systems can be reached through scheduled exports or SFTP. Anything without a connector is built as part of your engagement, and access always follows your permissions. See the platform.

It flows in automatically once a source is connected, and it keeps flowing, so the Brain works from current information. Connecting a source doesn't mean everyone sees everything; each person is answered from what they are already allowed to see. You can also drop in a file at any time.

It takes actions: updating records, drafting proposals, preparing follow-ups, checking invoices, and moving work between systems. Anything that matters waits for a person to approve it, and every action is logged.

Because the Brain is model-agnostic, a better model plugs into the Brain you already have. Your data, memory, and workflows don't change when the model does. We test new models against your own work before switching anything over.

• 92% less sales-team time on RFQs at Regal Technology Partners, a Tier 2 defense manufacturer.
• 5 months from first idea to a Brain built and launched across a 200-person manufacturer.
• 65% lower cost per AI task through model routing.
• 33.5% average increase in employee productivity within six months.

See customer stories.

Against a baseline we take with you in month one. For each workflow we measure the before and after: hours spent, cycle time, cost per task, error and rework rates, and how many people use it each week. You see the numbers in your progress reports, and the impact calculator gives you a planning estimate before you start.

We have built and run Brains for companies in defense manufacturing, specialty chemicals, energy field services, building technology, healthcare, financial services, media, and professional services. We also run nBrain itself on one.

Yes. Once we have both confirmed there is a fit, we can introduce you to a client in a similar industry. You can also read the Regal Technology Partners story and watch the film.

nBrain was founded by Danny DeMichele and Cary Johnson, two operators who spent decades scaling businesses before they built AI. We are headquartered in Carlsbad, California. We build like owners, because we are.

The founders stay close to every engagement. You also get a named team for the build, and your owner has a direct line to us.

Yes. Our sales, proposals, client reporting, CRM, task tracking, marketing, and delivery all run on the same platform we build for clients. Watch it work.

Book a strategy session. In one conversation we map where a Brain would give your people the most time back, which deployment fits your data, and what it would take. From there we write a proposal with the scope, timeline, and price. Book your AI strategy session.

Ask nBrain's agent at the bottom of this page. It answers questions about security, cost, timelines, and how a Brain would work for your company. Or email us at hello@nbrain.ai.