All twelve workflows from the DIFC setup guide, built as fixed rules that run inside your browser. The rules were written in advance with Claude and checked by a person. No AI service is called while you use it, so no client data reaches any model. Where a job needs an AI draft, the desk does the rule-based part here and the guide shows what the firm's own offline server adds.
connect-src 'none'. Even a bad script could not send data out.Same numbering as section 5 of the guide. browser means the whole job runs here. browser + server means the rule-based part runs here and the office server in World 1 adds the AI draft; the person who signs is the same either way.
| # | Job | Drop or type | Rules do, here | Person does | Runs | |
|---|---|---|---|---|---|---|
| 1 | Prospect brief | Public text about the prospect | Matches services to phrases, lists facts to verify, fills the intro email | Checks each fact, edits, sends | browser + server | Open |
| 2 | Onboarding pipeline | Pipeline export, one row per client | Stage board, days in stage, missing items, chase drafts | Adviser confirms, compliance signs the file | browser | Open |
| 3 | Fact-find and KYC | Every file for one client, plus a bank statement | Checks the nine required items, expiry and age, reads fields from text PDFs, scans source of funds | Compliance officer confirms every field | browser + server | Open |
| 4 | Meeting prep and file note | Holdings export before, raw notes after | One-page brief with questions to raise; file note with actions, owner, date | Adviser reviews before and after | browser + server | Open |
| 5 | Suitability report | Five fact-find answers and the model | Five consistency checks, then a draft from fixed clauses | Adviser edits and signs | browser + server | Open |
| 6 | Portfolio drift | Holdings export | Actual vs model by class, tolerance line | Adviser decides the rebalance | browser | Open |
| 7 | Equity and bond screening | Universe export with metrics | House rules applied, every pass or fail explained | Analyst signs the shortlist | browser | Open |
| 8 | Research notes | Report or factsheet PDF | Every figure with its page number, tick list | Analyst checks each figure, adds the house view | browser + server | Open |
| 9 | News watch | Headlines, one per line | Matches to client segments, rates impact by word list | Adviser reads the primary source | browser + server | Open |
| 10 | Client update emails | Client list with values and open actions | Fixed template filled per client, under 180 words, no forecasts | Adviser approves, sends from the mail desk | browser + server | Open |
| 11 | Staff follow-up tracker | Actions with owner and due date | Overdue, this week, later, by owner | Manager reviews overdue weekly | browser | Open |
| 12 | Audit log and verification | Nothing, it fills itself | Time, job, input, SHA-256 of the file, result, flags. Drop a file to prove it was the one used | Compliance samples monthly | browser | Open |
Paste public text about the prospect: their website bio, a press piece, a LinkedIn summary. The desk matches fixed phrases to services, quotes the evidence, lists every fact with a number or date for you to verify, and fills a first email. No source is fetched.
Drop an export with one row per client: Client, Stage, Stage since, then one column per document (Passport, Emirates ID, Address, Bank statement, Income, Source of wealth, Fact-find, Agreement, FATCA) holding Y, a date or blank.
Drop every file for one client. Each one is sorted by name and by the text inside text-based PDFs. Dates are read to flag expired IDs and old statements; passport number, Emirates ID and IBAN are read by pattern where the PDF has a text layer. Scanned photos are sorted by file name only. For the bank statement itself, use tab 3b.
Drop a statement export (CSV or Excel). The desk finds the date, description and amount columns, totals money in and out by month, lists the biggest payers, and flags large, cash, round and cross-border credits.
Before the meeting: drop the holdings export and the desk writes a one-page brief with the questions the numbers raise. After: paste the raw notes and it turns every commitment sentence into an action with an owner and a date.
Enter the fact-find answers and the model. The desk runs the consistency checks a compliance reviewer runs first, then fills a draft from clauses frozen in code. A mismatch is shown, never hidden.
Drop a holdings export (CSV or Excel) with a value column and either an asset class column or the holding name. Set the model weights and tolerance below.
Drop a universe export with a Name column and any of: Type, Market cap (bn), P/E, Dividend yield %, Debt/Equity, Rating, YTM %, Duration, Sharia (Y/N). Set the house rules; the table updates as you type.
Drop a research report or fund factsheet (text PDF). The desk pulls every sentence with a percentage or a money figure and tags it with the page number, so the analyst can tick each one before writing the house view.
Paste headlines one per line, or drop a text or CSV export from your news feed. Segments are your own keyword lists. Impact is a fixed word list, so it is a prompt to read the source, not a verdict.
Drop a client list with Client, Email, Value now, Value last quarter, Open action, Next review. Each draft is the same fixed template filled from that row: under 180 words, no forecasts, no promises. Copy it into the mail desk after you read it.
Drop an actions export with Action, Owner, Client, Due, Status. The actions from job 4 download in this shape. The desk sorts them into overdue, this week and later, by owner, for the manager's weekly review.
Fills itself. Every job you run on this desk is logged with the time, the input, the SHA-256 hash of the file and the result. Drop any file below to confirm it is byte-for-byte the one a run used. In memory only; download it before you close the tab.
F12 (Mac: ⌥⌘I), then the Network tab.