www.scale.net
PPS Intelligence · On-Premises AI

A private AI that knows your business — installed inside your walls.

Your yield knowledge, your HACCP program, your SOPs, and your live PPS production data — readable by your whole team through one assistant that never sends a word outside your building. AI is overwhelming; owning your pipeline shouldn't be.

On-premRuns in your facility
CitedEvery answer sourced
Read-onlyInto live PPS data
The Pipeline

Your knowledge + your live data morph into one assistant

PPS Intelligence holds your written knowledge in one hand and a read-only line into your live PPS database in the other. Questions that used to take an afternoon of pulling reports and flipping binders become one sentence.

Input · Yours

Your documents

  • SOPs & HACCP plans
  • Yield books & spec sheets
  • Excel & Word files
  • QuickBooks CSV exports
  • PDFs — even scanned binders
The Engine

PPS Intelligence

  • Understands your business and PPS — together
  • MCP tool-calling built in
  • Numbers computed by PPS, never guessed
  • Every answer cites its source
  • Open-source models — your data stays yours
Output · Answers

For your whole team

  • Yield-derived costing in seconds
  • Lot traceability on demand
  • Compliance answers, cited for the auditor
  • Reports in plain English
  • Onboarding that never gets tired
Plain English

Four ideas. No jargon.

01 / The engine and the fuel

Why your data is safe

The AI model is an engine. Your documents are the fuel. Fuel never becomes part of an engine: the assistant reads your material fresh each time and keeps nothing. It is never trained on your data — update a document, and it's current that minute.

02 / The librarian

How it knows your business

Imagine a librarian who has read every page in your filing cabinet — the yield book, the HACCP plan, every SOP — and never forgets where anything is. Ask a question; it pulls the exact page and tells you which binder it came from.

03 / The calculator

Why it never guesses numbers

Chatbots are famously bad at math — so ours doesn't do math. When a question needs a number, the assistant asks PPS itself to compute it, then explains the result. The arithmetic is done by the same system that runs your operation today.

04 / The amnesiac consultant

Why nothing leaks between uses

The engine is a brilliant consultant with a perfect briefcase and no memory: handed your file fresh for every question, retaining nothing afterward. Conversation history lives as ordinary records in your building — auditable, exportable, deletable.

Real Conversations

Ask it like you'd ask your best person

Document answers with citations, PPS-computed numbers, and read-only traceability — this is what day one looks like.

Your documents

What's our historical yield on boneless skinless breast from jumbo birds?

Your yield book shows 68–71% for jumbo BSB across the 2019–2024 sheets, with 69.4% the most common standard. Winter lots run to the low end — and the 2022 revision notes trimming-spec changes that moved it up half a point.

Yield Book · Poultry §4 · Sheets 2019–2024
Documents+ Live PPS

Raw ribeyes came in at $2.41/lb today. What should our yielded cost and list price be?

At your 71.3% yield standard, PPS computes yielded cost at $3.38/lb. Your margin policy for this class suggests a list of $4.12/lb — and two current price-list lines sit below that. Want the list?

⚙ compute_yielded_cost(ribeye) · Yield Book §2 + Price Policy
Your documents

Receiving just logged fresh poultry at 46°F. What does our plan require?

Your HACCP plan sets the receiving limit for fresh poultry at 41°F. At 46°F it requires: reject or segregate the load, record temps and times, notify QA, and document supplier corrective action — product may not enter storage until dispositioned.

HACCP Plan · Receiving CCP §3.2 — cited for the auditor
Live PPS dataread-only

Which customers received product from lot 2607103?

14 customers received 212 cases from lot 2607103 between 6/28–7/3 — full list with case counts and BOL numbers ready to export. Largest: 61 cases across 3 orders. A mock recall that took an afternoon takes a sentence.

⚙ trace_lot(2607103) · Your PPS database
Your SOPsen español

¿Cuál es el procedimiento para la limpieza de la sierra al final del turno?

Según su SOP de saneamiento: bloquee y etiquete la sierra, desmonte las guardas, lave con detergente aprobado, desinfecte, y registre en la hoja de saneamiento del turno. El supervisor verifica antes de energizar.

SOP · Sanitation §6 — same documents, your team's language
Live PPS data

Top 20 customers by margin this year, compared to last year.

Ready — 20 customers ranked by margin dollars YTD with last-year comparison. Headline: your #3 by volume is #11 by margin, and two mid-size accounts moved into the top ten on mix alone. Export to a spreadsheet?

⚙ run_report(margin_by_customer) — shows the exact query it ran

Illustrative conversations with representative figures — shown as the assistant behaves: cited, computed, and honest when the documents don't answer.

Privacy

Where your data physically lives

Your pricing methodology and yield knowledge are trade secrets. This system is designed around that fact — not with promises, but with walls. Open source is becoming necessary to take control of your data.

Public AI (cloud chatbots)PPS Intelligence On-Prem
Where the AI runsSomeone else's cloud, shared worldwideA machine in your building
Where your documents goUploaded to their serversNever leave your walls
Where conversations are storedTheir systems, their policiesYour building, your records
Works without internetNoYes — unplug it, it still answers
Is your data used to train AIDepends on their termsNever — architecturally impossible here
Who can see your yield bookUnknowableYour staff. Full stop.
Connected, Your Way

Local by default. Frontier when it counts.

PPS Intelligence is built local-first on open-source models — but it's not a closed box. It speaks the same tool-calling standards the wider AI world runs on, and can reach out to frontier models when you choose.

MCP tool-calling, integrated

The assistant works through the Model Context Protocol — the open standard for connecting AI to real systems. That's how it queries PPS, pulls documents, and runs reports: as governed tools with audit trails, not screen-scraping or guesswork.

It's also how the pipeline grows — new tools plug in without rebuilding anything.

MCPread-only PPS lineaudit loggedtool approval

Frontier connections — optional

Some moments call for the biggest brains available. When you want it, PPS Intelligence can route specific questions to frontier models from Anthropic or OpenAI — as an explicit, opt-in path you control, never a silent default.

Day to day stays private and local; the frontier is there when the right moment calls for it.

AnthropicOpenAIopt-in per useyou hold the keys
Models & Hardware

Open brains. Serious iron.

PPS Intelligence runs a variety of open-source models — chosen for your workload, swappable as better ones arrive. The engine upgrades; your knowledge base never moves.

The model lineup

NVIDIA Nemotron 30B-A3BFast · Agentic

The everyday workhorse — tuned for documents and tool-calling, streaming answers at a brisk reading pace.

NVIDIA Nemotron 120B SuperDeep Reasoning

The larger Nemotron for complex, multi-step questions — deeper analysis when the workload calls for it.

Qwen3.6-35B-A3BMultilingual · All-Round

A strong open all-rounder with excellent multilingual reach — the same SOPs, answered in the language your team works in.

GLM 5Flagship · DGX Station

The heavyweight — frontier-class open reasoning that the DGX Station runs comfortably.

The hardware we run it on

NVIDIA DGX Spark

The Appliance

A silent box the size of a thick cookbook, on a shelf in your office.

  • Runs Nemotron 30B-A3B beautifully
  • Comfortable for a moderate team through the day
  • Standard outlet, one network cable, shelf space
  • Draws less power than the TV in your break room
  • Graduates upward without rebuilding anything
NVIDIA Blackwell Station

The Server

A dedicated tower built on the NVIDIA RTX 6000 — the engine class we run for hosted clients.

  • Room for Nemotron 120B Super & Qwen3.6
  • Whole-company capacity; stays fast under load
  • UPS included; ventilated closet recommended
  • Headroom for every capability on the roadmap
  • The knowledge base moves with you — nothing rebuilt
NVIDIA DGX Station By Quote

The Flagship

Data-center-class compute at your plant — quoted per installation.

  • Runs GLM 5 comfortably
  • The deepest reasoning we can put inside your walls
  • For heavy multi-department, all-day loads
  • Scoped, built, and burned-in per site
  • Same rule as always: your data never leaves
Straight Answers

What it won't do — and why that's the point

A tool you'll trust for compliance and pricing has to be honest about its edges. These aren't fine print — they're design decisions.

Change anything in PPS on its own

Its database access is read-only by construction. Future action features execute only after a person approves, with an audit trail.

Guess a number

Yields, costs, and totals are computed by PPS itself. The AI requests the calculation and explains the result.

Invent a policy you don't have

If your documents don't answer a question, it says so — and points to the nearest related section.

Learn your data into the AI itself

The engine stays generic; your knowledge stays in your files. That keeps it upgradeable — and keeps your data portable and private forever.

Tiers & Fit

Many tiers. Custom everything.

Every plant is different — team size, document volume, how hard you'll load it. We offer multiple tiers and custom solutions that integrate with PPS; we'll scope the right fit together.

Connect

Frontier, connected

Start with frontier models from Anthropic or OpenAI connected to your PPS — the fastest way in, with your keys and your controls.

  • Frontier model access
  • PPS integration via MCP
  • Lightest footprint
Private

On-prem appliance

Your knowledge base and assistant running on hardware inside your walls. Data never leaves the building.

  • Fully local, open-source models
  • Knowledge base built from your binders
  • Serviced like your scales
Scale

Dedicated server

The full engine — whole-company capacity, the largest local models, and headroom for everything on the roadmap.

  • Deepest analysis, many users
  • Grows without rebuilding
  • Optional frontier routing on top
Don't see your shape?

Multi-site, hybrid, staged rollouts — we build custom solutions around PPS every week.

Scope it with us

Your knowledge, working. Your walls, keeping it.

Bring us one binder and ten questions. We'll show you the rest.

Start the conversation