No bank connection, and yet no tedious data entry: your statements, your AI (ChatGPT, Claude, Mistral, Gemini…), one file. Patrice shows you what changes before writing anything.
A blank, documented JSON file: it describes every collection and every field, with the allowed values. It is the same file as in the app, generated from its catalogue.
The instructions your AI must follow, in your language: invent nothing, respect the value lists, leave blank rather than guess, return the JSON alone.
The template, the prompt and your documents — bank statements, tax notices, loan schedules, brokerage statements, valuations. It returns the filled-in file.
Settings → Import → JSON file. Patrice compares the file with your account and shows every new, modified or missing line; you tick, it writes. Never blindly.
In this page's language. The template holds no data: one example owner, examples under "__examples" (never imported) and the full schema.
patrice-template-en.json — the same file as the "Download the template" button in the app's Import tab.
Download the template (JSON)It is also inside the file, under "__instructions": an AI that reads the file finds it on its own.
Patrice — personal-wealth data file (format patrice.v2, schema 2.2, prompt 2.1). App: https://patrice.app · Website and step-by-step guide: https://mypatrice.com/en/fill-in-with-your-ai/
You are an assistant that fills in a personal-wealth data file in JSON format for the Patrice app.
I am giving you: (1) this JSON file, whose "__schema" block describes every collection and field, with the allowed values ("enums"); (2) my documents (bank statements, tax notices, loan schedules, brokerage statements, valuations…).
Your task: fill the "owners", "assets", "budgets", "savingsGoals", "wealthGoals" and "taxProfile" collections from my documents.
Rules, in order of importance:
1. Never invent. A field my documents do not give stays absent — no estimate, no placeholder.
2. Never change an existing "id" and never create one for a new item (except a temporary negative number when another new item must reference it, e.g. a position inside an account through "parentId", or a linked loan through "linkedDebtIds").
3. Strictly respect the value lists in "__schema.enums" (asset types, currencies, frequencies…). If nothing fits, pick the closest type and explain it in "details".
4. "value" is the PURCHASE amount (or the capital borrowed), positive. Today's value goes in "revaluationValue" + "revaluationDate".
5. Dates as YYYY-MM-DD, months as YYYY-MM, amounts as plain numbers without symbol or thousands separator.
6. Owners ("owner") must exist in "owners"; add them if needed, with a short stable id.
7. Delete nothing. Do not touch fields marked "readonly", nor keys starting with "__".
8. Answer with the JSON only — no comment, no surrounding text, no Markdown block — and put in it ONLY what you add or change: the collections you touch, and within each only the items concerned. An existing item is given by its "id", its required fields (for an asset: "type", "name", "owner", "value", unchanged) and the fields that change; everything else — history included — is left out and keeps its value. Keep "meta" as it is, and DO NOT copy back "__schema", "__instructions" or "__examples": they are instructions for you, not data. A partial file beats a whole one copied out: it cannot be cut short on the way, and it erases nothing.
Before answering, check: every "type" appears in __schema.enums.assetType; every date is written YYYY-MM-DD; every amount is a bare number, no symbol and no thousands separator; every "owner" exists in "owners"; no existing id has changed; the JSON closes properly.
If a document is ambiguous, write what you understand in the item's "details" rather than deriving a number.Every collection is described in "__schema": the fields, their units, the allowed values (asset types, currencies, frequencies). Your AI has nothing to guess about the format.
You, your partner, your children — every asset belongs to someone.
Your companies and their partners, for pro-rata consolidation.
Accounts, savings, securities, property, loans, crypto… with the value history.
Income and expenses, month by month.
What you set aside, for what, by when.
The net worth you aim for in a given year.
Marginal rate, tax residence, rental regime, family situation.
A field your documents do not give stays absent: no estimate, no default. A doubt goes into "details", not into a number.
An existing "id" never changes; a new item gets none, except a temporary negative number to link two new items together.
Asset types, currencies, frequencies: those of the schema. If nothing fits, the closest type, explained.
The complete file, no comment or text around it — that is what the import expects.
You choose the AI and what you give it. Patrice only sees the file, at import time, with the diff in front of you.
Prefer an AI whose retention policy you know, and mask what is not needed — account numbers, addresses.
No bank connection, and yet no tedious data entryYour documents never go through Patrice
Create your free account, or try the demo first — the file is waiting in Settings → Import.