Private AI · Atlanta · Twilight Tech LLC

AI your client files are allowed to touch.

A private AI appliance on your network that answers from approved documents, cites its sources, and keeps third-party cloud out of the path.

Built for confidentiality-bound teams—law firms first, then any practice where client files, contracts, records, or internal knowledge cannot casually leave the building.

Try the live demo Discuss the Private AI Path
On-premisesYour hardware, inside your network
Cited answersEvery answer points back to its files
Team accessPhones and desktops, no cloud accounts
// The appliance

A box in your closet. An AI that answers from your approved documents.

A quiet appliance runs a local AI model and a private search layer over the document estate you approve. Your team asks questions, drafts updates, finds facts, and checks sources without sending the matter to a third party.

PRIVATE-AI//APPLIANCE
On your network
The metalA quiet, client-owned box sized to the number of users, documents, and models your team actually needs.
The brainA local model plus retrieval over approved matters, contracts, records, and SOPs.
The proofAnswers include source receipts so staff can inspect the file, page, and supporting excerpt.
The careSystem Care & Improvement covers updates, backups, monitoring, retrieval tuning, model upgrades, new document collections, and a human who answers.
⟵ your network boundary · no client data crosses this line ⟶
Built for confidentiality-bound document workLaw firms are the clearest first use case, but the architecture fits any practice where confidentiality is an obligation rather than a preference.
Replace shadow AI with sanctioned AIYour staff is probably already experimenting with public chatbots. Give them a tool that knows the work and fits the confidentiality rules instead of relying on bans nobody can enforce.
Useful scope, honestly drawnIt excels at answering, drafting, summarizing, and finding information inside your documents. It does not replace legal judgment, research subscriptions, or a human reviewing the work.
Everyone else writes policies around the cloud. We remove the cloud. The confidentiality claim is architectural: the model, index, and documents live on hardware you control. Each engagement also includes a short AI-use policy written to match the system your team receives.

Need AI to run coordination work instead? J-Bot Operations handles email, scheduling, tickets, reporting, and other cross-tool workflows behind approval gates.

See J-Bot Operations →
// Live demo · fictional legal matter

Ask a matter file. Get an answer with receipts.

This is the appliance experience in miniature. The room searches only its allowed matter packet, cites the files it used, and refuses to invent support that is not there. The documents and client are fictional; the retrieval and guardrails are real.

Bounded contextIt can search this matter packet—not the open internet or your real files.
Cited answersOpen a source chip to inspect the page, section, and supporting excerpt.
Cite or refuseAsk it to promise a win or ignore its boundary and watch the guardrail hold.

Want this on your own document estate? The Private AI Readiness Audit maps the files, permissions, hardware, and highest-value workflows before you buy a box. The fee credits toward a build.

Discuss Your Document Boundary
// Where it fits

For work that cannot casually leave the building.

Private AI makes sense when your staff repeats document-heavy work all day and the source material carries professional, contractual, privileged, or regulatory confidentiality obligations.

First wedge

Law firms: matters, discovery, intake, and client updates

Find chronology, compare records, draft plain-English updates, summarize depositions, and point every answer back to the matter packet.

Professional practices

Contracts, policies, records, and repeat questions

Give advisors, CPAs, and specialist teams a private knowledge room built from the files they already use to serve clients.

Regulated environments

Approved context with visible boundaries

Control which document collections enter the room, who can ask questions, and what evidence supports the answer.

Not every team needs a box. If your documents are already safe in an approved cloud system, your workload is light, or a simpler tool solves the real problem, the audit should say so. The goal is the right architecture—not hardware for its own sake.
// The engagement

Audit first. Prove one workflow. Then size the box.

No appliance quote should be based on vibes. We start with the document estate, permissions, users, workflows, and risk boundary—then build only what the evidence supports.

Start here
Private AI Readiness Audit
90 minutes · written report
Scope first working session + written report

A recorded working session and an engineering report that turns your files, users, risks, and best use cases into a buildable plan.

  • Document-estate and permissions map
  • Ranked workflows by value and risk
  • Hardware and deployment recommendation
  • Build estimate and 30-day path
Book the Audit
Validate next
Private AI 90-Day Pilot
one collection · one workflow · up to five users
Validate one controlled 90-day pilot

Test the highest-value use case inside a controlled boundary before committing to the permanent appliance.

  • One organization and one controlled environment
  • One approved document collection and workflow
  • Initial ingestion, retrieval, citations, and refusal behavior
  • Up to five users and 20 benchmark questions
  • Training, weekly quality checks, and final report
  • Pilot fee credits toward an appliance signed within 30 days
Scope the Pilot
Private AI Appliance
scoped after the audit
Build sized after the evidence

Client-owned hardware, local model, retrieval over the approved document estate, citations, deployment, and team training.

  • Hardware sized to the actual workload
  • Private document ingestion and retrieval
  • Role-aware access and answer receipts
  • Launch, training, and initial tuning
Discuss the Appliance
System Care & Improvement
month-to-month maintenance, tuning, and support
Improve ongoing system care

The appliance remains a maintained system instead of becoming an abandoned box in a closet. Care is month-to-month after launch; no automatic 12-month lock.

  • Updates, backups, and health monitoring
  • Retrieval tuning and answer-quality reviews
  • Model upgrades and local performance checks
  • New document collections and workflows
  • Quarterly improvement review
  • Priority human support
Ask About System Care
01 · ObserveMap the real work

Files, permissions, users, repeat questions, and risk.

02 · PilotProve one workflow

Ninety days, one approved collection, visible quality checks.

03 · BuildInstall inside the boundary

Hardware, model, retrieval, access, and training.

04 · CareKeep it useful

Measure quality, update safely, and expand deliberately.

// Why Twilight Tech

One accountable builder across the whole stack.

The builder

AI engineering plus the IT work required to make it real

I'm James Destrades Jr, the Atlanta-based founder of Twilight Tech. I built J-Bot because the standard AI tools did not fit the way a small operation actually works.

This product needs both disciplines: selecting and racking the hardware, configuring the network, building retrieval, drawing permissions, tuning the model, and supporting the people who use it.

The honest boundary

Private does not mean magical

Local models are excellent at reading and working from your approved documents. They may be weaker than frontier cloud models at broad, open-ended reasoning.

I will show you that line before you buy. The appliance should handle the work it can support with evidence and refuse the rest—not bluff its way through professional decisions.

// FAQ

The questions careful teams actually ask.

Does any client data leave our network?
In the on-premises design, the approved documents, retrieval index, and local model remain on your hardware. Remote maintenance is scoped separately and can be disabled entirely for fully offline environments.
What can staff use it for?
The strongest uses are answering from files, finding facts across a matter, drafting from approved context, summarizing records, comparing documents, and producing client-friendly updates with source receipts. It does not replace professional judgment or guarantee an outcome.
What documents can it read?
The audit identifies the useful and permitted collections. Typical inputs include PDFs, Word files, matter exports, policies, contracts, intake records, and internal procedures. Access should follow the same matter, department, and role boundaries your team already uses.
What happens when the internet goes down?
The core private document room continues answering because the model and index are local. Optional services that depend on the internet pause until connectivity returns.
Why start with an audit?
Because user count, document volume, permissions, workflow value, and model size determine the right build. The audit prevents an oversized box, an undersized promise, or an appliance where a simpler tool would have been enough.

Want AI your client files are allowed to touch?

Start with the live demo, then bring the real document boundary to a Private AI Readiness Audit. You'll leave with a concrete answer—even if that answer is not to buy the box.

Discuss the Private AI Path → Looking for workflow automation? Explore J-Bot Operations.