The Buyer's Guide · Enterprise AI · Your Terms
Modern AI for the work the cloud was never allowed to touch.

Everything a buying team needs in one place: who this platform is for — and who it isn't — what to require from any vendor, how to evaluate without taking anything on faith, and what the decision means for every seat at the table.

12-minute read Printable · Save as PDF Edition 2 · June 2026
01If you read nothing else

The whole case, in five claims.

Everyone already uses AI — the question is on whose terms. Sensitive questions go to public tools you don't govern, or internal bots guess with no source to check. A sanctioned, recorded way to use AI beats a ban no one follows.
LamFia is one app, every AI tool your team needs, on one foundation. Chat, writing, document production, instruction management, governed helpers, workspace-wide search, describe-and-draw diagrams, and an operations console — eight tools on the same approved documents, the same prompts and agents, and the same record, with a Workspace hub for setup, in-app help, and a named workspace per project. Switching between them is a single keypress.
Two promises most platforms can't make. Your data stays inside your walls — local-first, and in AirTight the software itself blocks every outbound call. Your content stays yours too — everything is plain Markdown on the open LGMD standard, so you're never locked out of your own work. One switch swings the platform between air-gapped and cloud-augmented; five routing profiles decide which engine answers each job, and PII analysis is pinned local in every one.
Every action lands on a tamper-evident record. Questions, drafts, helper runs — recorded with their sources, linked so any change is noticed. The position you take is "we can show what the AI did."
The decision is de-risked by design. A two-week pilot on your own hardware with one real workflow as the test; plain, portable files on an open standard mean walking away costs you the tooling, never the work.
The bottom line: if some of your work can't live in someone else's cloud, this platform was built for exactly that work — and you can verify every claim above in a demo on your own kit.
02The one question that sorts everything

Can your data live in someone else's cloud?

Most AI products assume it can. For a great many serious organizations — wherever the material is privileged, regulated, or simply must not leave the building — it cannot. That single fact reframes the whole conversation, and the tree below sorts it in under a minute.

Fig. 1 — Where does each kind of work belong?Decide per workflow, not per company
NO YES YES NO YES NO START · FOR THIS WORKFLOW Can this work's data live in someone else's cloud? Must you prove zero egress — not just promise it? Still need answers from your documents, on a record? ZERO EGRESS · ENFORCED AirTight setup Fully offline. The software blocks every outbound call — and proves it. THE DEFAULT LamFia · local-first AI runs on your machines; cloud engines stay opt-in, job by job. USE BOTH LamFia + cloud engine Open work on a cloud engine, while your documents and the record stay home. THE HONEST ANSWER Cloud AI, for that job Fine for open work. Bring the protected work back to LamFia. WHICHEVER BRANCH — THE PROTECTED WORK HAS A HOME, AND THE RECORD STAYS YOURS

Run each important workflow through the tree. Most organizations land on two or three branches at once — that's normal, and the platform serves all of them.

LamFia is a strong fit if…

  • Some of your work can't leave the building — privileged, regulated, classified, or simply too sensitive for someone else's cloud.
  • You'll be asked to prove what the AI did — by an auditor, a regulator, a court, or your own board.
  • You want one governed platform, not a scatter of per-team AI subscriptions with no shared record.
  • The people who know the work aren't engineers — and you want them building the automation anyway, in plain words.

It probably isn't if…

  • All your work can live in a cloud and you mainly want the largest ecosystem — a cloud assistant may serve you well. Use LamFia alongside it if protected work appears.
  • You need rule-based automation across business systems — a dedicated automation tool reaches further than an AI workbench should.
  • The AI platform itself is your product — then building your own may be the right call. Buy the plumbing only if it isn't.

Wherever you land, you also choose where LamFia runs — and the options share the same restraint as everything else: the simple, signed desktop install stays the default, and each step onto your own infrastructure is available, optional, and on your terms.

The engine on your own servers

Run the AI engine as one sealed, signed, self-contained package on hardware you own — quick to stand up, identical on every machine, and able to be sealed shut from the internet entirely: the strongest form of the AirTight promise. Built on open, vendor-neutral building blocks, so adopting it ties you to no single container vendor. Available and optional; it packages the engine, not the desktop; large-scale cloud hosting, managed databases, and high availability are a deliberate later release.

The backend as an enterprise service

For a platform team, run the Mission Control backend as a connected service inside your own cluster — on your identity provider, your database, and your monitoring — so no customer data goes to a vendor cloud, and a self-managed setup carries no copyleft obligation. Off by default; it requires enterprise configuration; this is the connected story, not the air-gapped one (that's the desktop and the sealed package).

03Why this is one platform, not a stack of subscriptions

The six value pillars.

The tools people actually want

Chat, writing, documents, automation — the familiar experiences everyone already reaches for, made dependable for company work. Useful on day one, no training course.

One foundation, one truth

Every tool reads the same approved documents and writes to the same record — the answer in chat matches the draft in the writer.

Your experts build the automation

Instructions, helpers, and document templates are built from plain words — by the people who know the work, not engineers.

It runs on your machines

The AI works on your own computers by default. In the AirTight setup, the software itself blocks every outbound call — enforced, not promised.

A record you can hand an auditor

Every action recorded with its sources and linked so any change is noticed — “we can show what the AI did.”

Costs you can see — and no lock-in

Per-engine cost tracking, shared capacity, one licence that never bricks — and your work stays in plain, portable files on an open standard (LGMD).

04One app, the tools you need

The eight tools at a glance.

Fig. 2 — One application · eight tools · one foundation · one recordNo copies · no migrations
One shared foundation Trusted documents · approved instructions · governed helpers · your AI engines
Every action, from every tool, on one tamper-evident record
ToolWhat it doesThe AI concept (plain English)
AC AI ChatAnswers your questions from your own documents, with sources shown.Answers from your documents (RAG)
SW SmartWriterDrafts content and keeps it safe in one place.Intent + strip personal details
PS PromptStudioBuilds, scores, and approves AI instructions.AI instructions, scored
SX StrataryxWrites long documents, section by section.A coordinated team of AI agents
AA AI AgentsReusable, governed helpers your teams can trust.Specialised agents, scoped
AS AI Powered SearchFinds anything across your whole workspace, then opens it where it lives.Words and meaning, merged (top-tier extras off by default)
DG DrawGenieDescribe a diagram in plain words and it's drawn — on your own machine, yours to keep.Generation from a plain-words description
MC Mission ControlRuns it all like a managed service.Engines, the record, sharing

Each chip opens that tool’s full page — or start at the eight tools and drill in.

05The trade-off the platform removes

Capability, control, cost — the old rule was “pick two.”

Capability modern AI · eight tools Control your data · your record Cost visible · predictable LamFia ALL THREE
  • Cloud AI gives capability — and takes control: your data leaves, the record isn't yours.
  • Build-it-yourself gives control — and takes cost and time: you assemble and maintain the plumbing.
  • Doing nothing protects the budget — and forfeits capability while shadow AI grows unrecorded.
  • LamFia is built so you stop choosing: modern AI on machines you own, with the record and the budget on your side.
06How it compares · honest positioning

Named honestly, not bashed.

LamFia isn't competing to be the best cloud AI. It's the platform for the work that can't go to a cloud at all — and for proving what the AI did afterwards. Often the honest answer is "use both."

The alternativeWhere it's genuinely goodWhere LamFia wins
A cloud office AI assistantDeep integration with everyday apps, huge reach.Runs with nothing leaving the building, draws only on your approved documents, keeps the record on your own disk.
A frontier-model providerThe most capable cloud engines available.The governed workbench around the engine — you can even point LamFia at one while your documents stay home.
Build it yourselfFull control; capable open building blocks exist.The record, sharing, governance, migrations and tooling are already shipped and maintained. Buy the plumbing; build your differentiator.
Doing nothingNo project, no cost — for now.A sanctioned, recorded way to use AI is safer than a ban no one follows — and it's available now, on hardware you own.
If your reality is…The honest position
"We want the best cloud model, and our data can live there."A cloud AI may suit you; use LamFia for the work that can't go there.
"Some of our data can't leave our walls."This is exactly LamFia's world — and you can still point it at a cloud engine for the rest.
"We're thinking of building it ourselves."Buy the plumbing, build your differentiator — LamFia already ships the hard parts.
"We're not doing AI yet."A governed platform beats an unenforceable ban; start with a low-risk pilot.
07The evaluation checklist

What to require from any platform — including ours.

Take this list into every vendor conversation. Each row is a requirement worth insisting on, why it matters, and the question that separates a real answer from a brochure. LamFia is built to pass all twelve — in a demo, on your own hardware.

RequireWhy it mattersAsk the vendor
Runs on hardware you ownIf the AI only works in their cloud, your data's location is their decision, not yours."Show it running with the network cable unplugged."
Provable zero egress, when neededA policy promise is not a control. The product itself should enforce the boundary."What happens if the offline guarantee fails — does the software refuse to start?"
Answers grounded in approved documentsUngrounded bots answer confidently and wrongly — with no source anyone can check."Show me the exact paragraph this answer came from."
A tamper-evident record of every action"We used AI" is not an audit position. "We can show what the AI did" is."Export the record of this session and show me how a change would be noticed."
Human-approved handling of personal detailsSilent redaction is a risk of its own; nothing should be removed without sign-off."Who approves each removal — the software or a person?"
Governed instructions & helpers, with lifecycleA sensitive automation must not quietly change. Versioning and approvals are the control."Walk me through draft → approved → retired for one helper."
Built by domain experts, not just engineersIf automation requires code, the people who know the work can't build it — and won't use it."Have a non-technical person build a working helper in front of us."
Open, portable formats — no lock-inWalking away should cost you the tooling, never the work."Open one of our files in an ordinary editor, right now."
Cost visibility per engine and jobAI spend without a dashboard becomes a surprise at quarter end."Show me what this morning's work cost, engine by engine."
Degrades gracefully — never bricksA licence lapse must not take your tools or your work hostage."What exactly stops working when the licence lapses?"
Finds across everything you have — by word and by meaningKnowledge scattered across tools is knowledge you can't reuse; one search by meaning and by tag turns it back into an asset."Search our whole workspace from one box — and show that the default search runs locally."
Runs on your own identity, data, and cluster when you need itAn enterprise platform team needs to run AI the way it runs everything else — its own sign-in, its own database, no vendor cloud holding the data."Can the backend run in our cluster, on our identity provider and database, with nothing going to your cloud?"

Print this page and bring it to every demo — ours included. A vendor who hesitates on any row is telling you something.

08Trust — why you can say yes

Trust is a property of the system.

The AirTight promise

The fully offline setup blocks every outbound call and refuses to start if the block isn't working — privacy as a property of the product, not a line in a policy.

Every action on the record

A question, a draft, a helper run — recorded with sources, linked so any change is noticed, ready to hand an auditor.

No lock-in

Plain, portable files on a published, vendor-neutral standard. If a licence lapses, the software keeps working with basic features. It never bricks.

Local-first, your choice per job

The AI runs on your hardware by default. Reach for a cloud engine only where your policy allows — and in AirTight, never.

09Proof — regulated teams, sensitive work

The pattern repeats.

People want modern AI, the data can't leave the building, and a grounded, on-machine, recorded tool is what finally makes it possible. Three anonymised cases — no invented numbers.

Fig. 3 — What changed, before → afterAnonymised · qualitative
Clause drafting from confidential precedentFirst AI tool their ethics committee approved on first review
Hours per clause
Minutes
Large proposals inside a secure perimeterFully offline · approved on the first review
Days of writing
A working session
Reviewing AI instructions, governedReplaced several subscriptions with one platform
Once a quarter
Every week
Internet traffic in AirTight modeMeasured at the network boundary
“Trust us” — varies
None — verified

Bars are qualitative, drawn from the cases above — the point is the order of magnitude, not a benchmark.

10The budget conversation

What it replaces — and what you can finally see.

This isn't a price list — pricing depends on your estate, and that conversation belongs in the pilot. What a buying team can establish now is the shape of the cost: what the platform consolidates, and what becomes visible that today is guesswork.

Fig. 4 — The consolidation, and the visibilityShape, not a quote
Today · a typical scatter
Chat subscription, per seat Writing assistant, per seat Prompt-management service Document automation tool Agent platform Untracked shadow-AI use
One platform · one licence
  • Per-engine cost tracking — the AI budget becomes a dashboard
  • One strong machine shares capacity across a team
  • Never bricks — a lapse degrades features, not your work
  • The record consolidates too: one trail, not six

One regional bank's instruction-review case above replaced several subscriptions outright — your mix will differ; the pilot makes it concrete.

11For every seat at the table

The question each stakeholder will ask — answered.

Enterprise decisions are made by a committee, not a person. Forward each card to its owner; each links to the section with the full argument.

Executive sponsor
"What does the organization actually get?"
The AI your teams already want — answers they can check, drafts from your own material, big documents in a session, recurring work automated — without the exposure of ungoverned tools. And a two-week, walk-away pilot instead of a leap of faith.
The summary · 01 →
Security & compliance
"Does data leave, and can we prove what happened?"
Local-first by default; in AirTight the software enforces zero egress and refuses to start otherwise. Every action lands on a tamper-evident record with its sources — exportable, ready for an auditor. When you run the enterprise backend, it sits on your identity provider, your database, and your monitoring, with no customer data going to a vendor cloud and — self-managed — no copyleft obligation; the sealed server package can carry a signed, checkable list of what's inside.
Trust · 08 →
IT & AI leads
"What are we signing up to run?"
One application, eight tools, one console. Engines discovered, routed, and costed in Mission Control; no agent framework, record store, or governance layer to assemble — the hard parts ship maintained. Run it where you say: the simple desktop install, the engine as a sealed package on your own servers, or the backend as a connected service in your own cluster on your identity and data (off by default; large-scale cloud hosting is a later release).
The tools · 04 →
Finance & procurement
"What does it cost — and what if we leave?"
One licence consolidates a scatter of per-seat subscriptions; per-engine cost tracking makes spend a dashboard. No lock-in: open, portable files — and the software never bricks on a lapse.
The budget · 10 →
12The jargon, translated

Twelve terms, in plain words.

Every AI pitch leans on these. Here is what each means for your work — with the industry term, so you'll recognize it in other vendors' materials.

Answers from your documents
The AI looks inside your approved documents first and answers from what it finds — with the source shown, so it stops making things up.
Industry term · RAG
Search by meaning
Find "ways to reduce churn" even if the document says "customer retention" — search understands what you mean, not just your words.
Semantic search
The AI engine
The model that reads and writes. Different engines suit different jobs — fast vs deep, local vs cloud — and you choose per job.
Large language model
AI instructions
The brief you give the AI before it answers — written once, carefully, like briefing a new colleague. In LamFia they're scored and approved.
Prompts
AI agents
Packaged helpers with one job each — researcher, reviewer, writer — each with its own rulebook and a short list of allowed tools.
Agents
Tools the AI may use
Small, named capabilities you allow a helper to use — look up a figure, open a document. Each request is checked and recorded.
Function calling · MCP
Sharing across machines
Two offices lend each other AI capacity without moving files — each keeps full control of its own documents and rules.
Federation
The tamper-evident record
Who asked, which documents were used, what came back, when — linked so even a small change is noticed.
Audit trail
The open standard
Your documents and instructions stay plain, portable files that open in any editor — leaving costs you the tooling, not the work.
LGMD
Search that blends exact words with meaning
One ranked list that merges word-for-word matches with matches by meaning, so a half-remembered query still finds the right item — and the default ranking runs on your machine.
Reciprocal-rank fusion
A guaranteed shape
You write the worked example once; the result always comes back with the same sections and headings, rather than whatever the AI engine returns on its own.
Output contract
A stateless service
A backend that keeps nothing between requests — each request stands on its own — so it runs in your cluster like any other service, scales, and survives a node failing.
Stateless service
13The de-risked decision

A short pilot, on your data, on your hardware.

You don't take any of this on faith. The decision becomes "did this workflow get better on our own kit?" — not "do we trust the pitch?"

Day one
Install

Set it up and bring your own documents in.

Within days
Run one real workflow

The test that actually matters to you.

At two weeks
Review together

Look at the outcome side by side.

Decide
Expand — or walk away

If it doesn't help, keep all your data and walk away.