A new generation of spreadsheet formulas

Every cell,
A new possibility.

Every cell is a workload. Connect data, files, code and AI through formulas. Reuse each result—and see exactly where a workflow goes wrong.

Designed to run locally or across approved compute nodes, through the spreadsheet or its APIs. Work individually or together, turning data you choose to share into small, reusable models for everyday tasks and enterprise workflows.

Sample formulas Request private access

Local by design. Cloud by choice. Errors remain visible.

From input to result

B2 CogSheet^™ cellYour formulaInput → OutputResult or #ERR!

“Put the product specifications into columns.”=WEB.EXTRACT(A1, "product specifications", COLUMNS(B1:G1))

A web page → structured rows you can compare.

Explore the full product direction. These illustrate planned adapters, not live connections.

Every cell is a unit of work

One cell. One job.
A result you can use.

01 · InputA number, a file,
another cell

02 · FormulaCalculate, read,
predict, run AI

03 · OutputA result.
Or a visible error.

For everyday life. For the work you do.

Make everyday life
a little easier.

With the things you already use: a URL, a social feed, a photo or a voice message from a friend. Give a cell an input, describe the job and use the result in your next step. The same idea scales from a household list to a business workflow.

01 · A link

“Compare these offers.”

A shop page, a supplier catalogue or a research source. Bring the facts you need into columns.

02 · A social feed

“What keeps coming up?”

Turn a permitted feed or saved export into topics, recurring questions and useful signals.

03 · A photo

“Read this for me.”

A receipt, an invoice or a handwritten list. Extract the fields, then calculate with them.

04 · A voice message

“What do I need to do?”

A friend’s voice note or a team recording. Turn the words into a transcript and reviewable actions.

Already in this demo

100 receipts.
One small household micro-model.

From just 100 synthetic receipts, we built a small predictive model of shopping habits. It measures the gaps between purchases and uses them to suggest what may be due again.

That is a first step toward useful AI at home: a familiar question, answered from your own everyday data.

Try “What should I buy?”

INSIDE THE SAMPLE MODEL

Milk

Purchase observations: —
Typical purchase interval: —
Next purchase estimate: —

A deterministic purchase-cadence model. It estimates shopping needs; it does not measure what is left in the fridge. Training and exporting neural models are planned capabilities.

The input scenarios above describe the full product direction. This public demo uses prepared receipts and a bundled offers feed; personal uploads, live URLs and social connections are not enabled here.

Real outputs from synthetic data

Try a familiar question.

These answers are calculated from the sample household’s 100 receipts. Choose an example to load its formula, then press Run formula in the sheet. All public examples run without paid AI calls.

Live sample workbook

Every input. A next step.

This workbook is generated from a fixed seed and reference date. Every formula below runs in the CogSheet^™ deterministic engine and returns the same result for the same input.

E2fxReady

6 cells computedThird-party calls: 0Target runtime: ecOS >_ CogPOSIX

#ERR · know exactly where to look

If a step fails,
the cell tells you where.

A complex formula can involve many cells, including AI. When a step fails or its output fails a check, that cell shows an error such as #MODEL!. You can identify the input, formula or model result that needs attention.

Follow a problem through the cells.

Interactive illustration · fixed sample outputs, no AI call.

A2 · INPUT

Receipt photo

Original input stays available.

B2 · AI EXTRACTION

£17.99

Expected output: a numeric total.

C2 · CALCULATION

£22.99

=SUM(B2, 5)

D2 · RESULT

Ready to use

Uses the result from C2.

B2 supplies a valid number. C2 adds £5, and D2 can use the result.

Find the cause, even in a long workflow. The intended full-workbook behavior is to mark dependent cells as blocked until the failed step is corrected and run again. The illustration above demonstrates this on a small chain.

Checks make failures visible. Invalid formats, timeouts and detected inconsistencies can become cell errors. A plausible but incorrect AI answer still needs appropriate checks or human review.

Beyond the kitchen · full product direction

The same cells.
Whole new possibilities.

Keep the familiar rows, columns and calculations. Add formulas that process documents, audio, images, web pages, signals and models, using capabilities compatible with ecOS >_ CogPOSIX.

These are planned contract examples. Their syntax and adapters are still to be implemented; the public sheet above runs the pantry formulas today.

Ask for a formula in your own words

“Show bills with a missing due date.” The assistant proposes a formula for your cell. You review it, run it and use its result in the next calculation.

=TABLE.FILTER(BILLS, DUE_DATE = "")=SUM(C2:C12)

Web research into columns

Read an authorized web snapshot, extract only the requested facts and spill schema-validated rows into named columns with source evidence.

=WEB.EXTRACT(A2, "all product specifications", COLUMNS(B1:G1))=WEB.CITE(B2:G20)

Documents, OCR and fields

Preserve page coordinates while turning PDFs, scans, contracts and invoices into searchable text, layouts and reviewable structured records.

=DOCUMENT.OCR(A2, "en", "layout")=DOCUMENT.EXTRACT(A2, SCHEMA("invoice"), pages: "all")

Audio, speech and translation

Denoise a recording, transcribe it with timestamps, identify the most frequent meaningful terms, translate it or synthesize an approved voice output.

=AUDIO.DENOISE(A2, profile: "speech")=AUDIO.TRANSCRIBE(B2, "en", timestamps: TRUE)=TEXT.TOP_TERMS(C2, 20, stopwords: "en")=SPEECH.SYNTHESIZE(TRANSLATE(C2, "en", "es"), voice: E1)

Vision, video and inspection

Detect, segment or measure objects in images and sampled video frames while retaining the exact frame, region and model evidence behind each result.

=VIDEO.SAMPLE(A2, every: "2s")=VISION.DETECT(B2:B120, classes: {"defect","scratch"})=VISION.SEGMENT(C2, target: "damaged area")

Private knowledge and language

Build permission-aware indexes across text, images and audio, retrieve relevant evidence locally and generate summaries that remain linked to their sources.

=EMBED.INDEX(A2:A500, modality: "auto", placement: "local-required")=SEMANTIC.SEARCH("termination clauses", B2, 10)=LANGUAGE.SUMMARIZE(C2:C11, format: "decision memo")

Forecasting, signals and security

Forecast calibrated time series, detect unusual sensor patterns and classify security events without turning a model score into an automatic privileged action.

=SERIES.FORECAST(A2:A365, horizon: 30, interval: 0.95)=SERIES.ANOMALIES(B2:B10000, baseline: "weekly")=SECURITY.CLASSIFY(C2, profile: "event-v1")

Code and model workshop

Generate reviewable code capsules, reuse pinned environments, prepare labelled datasets and export validated small models for governed local execution.

=CODE.CAPSULE("python", A2, lockfile: B2)=MODEL.TRAIN(A2:F500, label: F, template: "classifier")=MODEL.EXPORT(G2, "onnx")

Your work. Your data. Your choice.

Designed to run
on your computer.

The goal is a complete offline experience on ecOS >_ CogPOSIX: your workbook, your data and locally runnable models, including small models built from your own examples. Sovereignty means keeping control of that work without a required third-party AI service.

The cloud hosts this preview. It does not define the product. This private beta lets you explore CogSheet^™ in a browser while the local integration is being built. The full product is designed to make online models and approved compute nodes optional, with a visible execution choice. Local-only work must never silently fall back to the cloud.

What does “deterministic” mean here?

Each cell declares its input, operation and expected output. Fixed inputs and deterministic operations give repeatable results. AI outputs need explicit checks; replaying an accepted result avoids asking a model the same question again. Fresh model runs are not automatically identical.

Input→Formula→Execution→Result or error

What comes next

From personal tasks.
To enterprise workflows.

Today: a working sample, visible formulas and a small predictive model. Next: your own files, reusable code and models, and execution on your own computer.

Durable workbooks

Per-user workspaces, signed uploads, artifact storage, asynchronous jobs, cancellation and execution receipts.

Local

ecOS >_ CogPOSIX bridge

Local capability discovery, signed manifests, offline replay and explicit local or hosted placement per cell.

Code

Bounded code capsules

Reviewable generated code, pinned dependencies and reusable isolated environments with no default network access.

Models

Small-model workshop

Dataset and label export, validated training templates and model-specific ONNX export where technically supported.

Fabric

Governed distributed execution

Approved-node placement, node identity, revocation, permissions and transfer accounting across a future trusted mesh.

Private preview

What would you put
in your first AI cell?

A personal project, a team’s workflow or an enterprise process. Request private access and help shape CogSheet^™.

cogsheet@logforce.com