⚙ Settings

Saved to this browser only — the local server uses it when you run an appraisal.

AI model · Sonnet 5.5 is the default
Loading available models…
All models run the same appraisal pipeline (web search, JAROT cross-check, comparable verification) at temperature=0 with adaptive thinking and prompt caching. Your choice is saved to this browser.
Prepared by · your details on reports
Pick your name and your details fill in automatically — shown in the "Appointed Broker" block of every report you generate, so clients contact you directly. You can still edit any field, or leave it all blank to use the office details. Saved to this browser.
Staff PIN Change app access PIN On the live site the PIN is set in the Render dashboard: Environment → VBAS_PIN_CODE, then redeploy. (A PIN changed here would reset on the next restart, so it's turned off.)

Report archive

Every report you generate is stored locally on this PC. Newer versions for the same business automatically replace the older ones — only the latest of each type is kept.

Stored locally — not uploaded anywhere. Tip: click Compare on two rows to see a side-by-side diff.

AI How the appraisal works

A quick technical map for Vic Brokers staff or anyone reviewing the appraisal logic.

01 · Inputs

Reads the whole brief

Uses business name, address, website extract, typed notes, and every uploaded supporting file. Text is extracted from PDF, Word, Excel, CSV and TXT files. JPG, PNG and WebP files are sent visually to the model; scanned PDFs are forwarded natively so Claude can read the text directly.

02 · Engine

Claude — Sonnet 5.5 or Opus 5.5

Every appraisal runs through the Anthropic API. Claude Sonnet 5.5 is the default — fast and well suited to everyday appraisals. Choose Claude Opus 5.5 above for deeper reasoning on complex or high-value appraisals; it is slower and costs about twice as much per appraisal. Both run with adaptive thinking enabled and prompt caching on for the JAROT chart, and every figure is cross-checked against the JAROT chart after the AI responds.

03 · Appraisal

JAROT-led pricing

The AI extracts evidence, normalised profit/PEBITDA, addbacks, source notes, likely JAROT industry and the closest EBITDA band. The server then checks the final low/mid/high range against the local JAROT chart so the visible appraisal reconciles to the applied multiple range times maintainable PEBITDA (or EBITDA for larger businesses and professional practices, or revenue for accounting practices, planners and brokers), with stock kept separate.

04 · Comparables

Market evidence is screened

Search is limited to Victorian listings on approved sites only: seekbusiness.com.au, commercialrealestate.com.au, australia.businessesforsale.com, anybusiness.com.au, bsale.com.au, business2sell.com.au and businessforsale.com.au. The app prioritises Seek Business and Commercial Real Estate first, targets 5-10 scale-matched listings, and removes dead, sold, generic, interstate or search-page URLs before reports are shown.

05 · Outputs

Reports and listing draft

Creates separate vendor appraisal and buyer business profile reports in DOCX/PDF, plus a concise draft listing ad. Reports are saved locally in the archive, replacing older versions for the same business/type/format. The main page keeps the latest appraisal until Clear appraisal or New appraisal is clicked.

06 · Support

Where to inspect issues

Runtime errors go to App Files\logs\app.log. Staff PIN, default model and API keys live in .env. Key files: app.py for AI calls and appraisal logic, jarot_data.py for JAROT bands, profile_doc.py and profile_pdf.py for report layouts.

Prompt rules in plain English

Use all supplied evidence, extract before appraising, do not invent facts or URLs, do not use irrelevant or out-of-scale comparables, prefer verified Victorian listings from the approved sites, show uncertainty through confidence and risk notes, keep stock at value separate, and let the server reconcile final pricing to JAROT.