Interactive demo — runs entirely in your browser

This is what AI inside FileMaker feels like.

Below is a working simulation of a FileMaker solution we might build for a consultancy. The familiar grey window is FileMaker. The amber panel is what we add. Ask the database questions in plain English, transcribe a meeting, pull structured data out of a scanned invoice — then open the drawer and read the actual script steps that drive it.

Straight up The data operations, workflows, and every FileMaker script step shown here are real. The AI inference is scripted in this browser demo — in production it runs on a local model on your machine, or Claude via MCP. Either way, no data leaves your control.

Harbour & Marsh CRM — hosted on your server

CompanyCitySector StatusRevenue YTDOutstanding

Recording

—

Client

—

Transcript

AI Summary

Action Items

Actions — via portal

No actions yet — transcribe a recording

Received

FENWICK MARINE SUPPLIES LTD

Unit 9, Albion Dockside Estate · Bristol BS1 6UT · VAT GB 384 2219 07

Invoice INV-2417 Date: 28 June 2026

Description	Qty	Unit	Amount
Marine-grade fastener kit, A4 stainless	12	£38.50	£462.00
Anti-fouling coating, 5L	6	£112.00	£672.00
Deck hardware inspection & certification	1	£850.00	£850.00

Subtotal	£1,984.00
VAT @ 20%	£396.80
Total due	£2,380.80

Payment terms: 30 days net

Supplier

Invoice No.

Date

Subtotal

VAT

Total

Terms

Line Items

Invoice Lines — via portal

Awaiting extraction…

AI Assistant 127.0.0.1 · local

Ask the 24-record sample CRM a question in plain English. The answers are genuinely computed from the data — watch the found set change.

# Ask the local model to translate a question into a find
Set Variable [ $q ; Value: GetAsURLEncoded ( Globals::Question ) ]
Insert from URL [ Select ; With dialog: Off ;
  Target: $json ; "http://127.0.0.1:8631/query" ;
  cURL options: "-X POST -d question=" & $q ]
Set Variable [ $find ; Value: JSONGetElement ( $json ; "find_request" ) ]
Perform Find [ Restore: $find ]
Set Field [ Globals::Answer ; JSONGetElement ( $json ; "answer" ) ]

Pick a recording. Transcribr turns it into a searchable transcript attached to the record — then the model summarises it and files the action items.

processing… — real-time

# One script step. That's the integration.
Insert from URL [ Select ; With dialog: Off ;
  Target: $json ; "http://127.0.0.1:8623/transcribe" ;
  cURL options: "-X POST
    -F \"file=@" & Meetings::AudioPath & "\"
    -F \"model=medium\"" ]
Set Field [ Meetings::Transcript ;
  JSONGetElement ( $json ; "text" ) ]

{
  "text": "Right — thanks everyone for making time…",
  "language": "en",
  "duration": 763.2,
  "processing_time": 54.1,
  "model": "medium",
  "segments": [ … ]
}

Extractr reads the scanned invoice on the left against a schema you define, and returns exactly the fields you asked for — as clean JSON.

// your schema, your field names

{
  "supplier": "string",
  "invoice_number": "string",
  "invoice_date": "date",
  "line_items": [ { "description", "qty", "amount" } ],
  "subtotal": "currency",
  "vat": "currency",
  "total": "currency",
  "terms": "string"
}

reading document… — confidence

# Send the container + schema, get structured JSON back
Set Variable [ $schema ; Value: Invoices::ExtractionSchema ]
Insert from URL [ Select ; With dialog: Off ;
  Target: $json ; "http://127.0.0.1:8641/extract" ;
  cURL options: "-X POST
    -F \"file=@" & Invoices::ScanPath & "\"
    -F \"schema=" & $schema & "\"" ]
Set Field [ Invoices::Supplier ;
  JSONGetElement ( $json ; "supplier" ) ]
# …one Set Field per schema key, or loop the JSON

From demo to production

Everything you just drove exists for real.

Each station in the playground maps to a shipping product or a project we've already delivered for clients.

Station 01 — Ask

Semantic search & natural-language queries

We've built this at production scale: 900,000 records searched in plain English with answers in 2 seconds, on a local LLM with no per-user licensing. The productised version is in development.

See the case study →

Station 02 — Transcribe

Transcribr

Speech-to-text for FileMaker as a single-file executable. Four modes, five models, any audio format, Core ML accelerated. One Insert from URL step and you're done.

Get Transcribr →

Station 03 — Extract

Extractr

Schema-driven document extraction. We've processed 10,000+ medical PDFs into structured FileMaker data for a genetics laboratory. Extractr is that pipeline, productised.

Extractr details →

/01

100% local by default

Our tools run on localhost. No cloud, no API keys, no data processing agreements — the data never leaves the machine.

/02

Native FileMaker integration

Insert from URL and JSONGetElement. No plugins, no middleware. If you can write a FileMaker script, you can use this.

/03

Trusted by the specialists

Claris Partner, AI ETS member, and AI advisor to Claris Platinum Partners. Speaker at Claris Engage, EngageEU, dotfmp and Rome FileMaker Week.

/04

Claude via MCP, when you want it

For conversational AI over your whole system, we connect Claude directly to FileMaker with MCP — within your existing security model.

Your system next

Imagine this panel bolted onto your solution.

A 30-minute call is enough to work out what AI could do inside your FileMaker system — and what it would take. No obligation, no sales pitch.

Book a Free Consultation Explore the products