Remember Finances

Import financial documents — bank statements, receipts, invoices — into Neotoma memory with structured transaction extraction and full provenance.

When to use

When the user has financial documents (CSV/PDF bank statements, receipt images, invoice PDFs) and wants to persist the transactions as structured, queryable records.

Prerequisites

Run the ensure-neotoma skill first if Neotoma is not yet installed or configured in your current harness.

Supported sources

Source	Format	Method
Bank statements	CSV, PDF	File read, parse_file for PDFs
Receipts	Images (JPEG, PNG), PDF	Vision, parse_file
Invoices	PDF, email attachments	parse_file, email MCP
Plaid	Live bank data	Plaid MCP (optional)

Workflow

Phase 0: Verify Neotoma

Confirm Neotoma MCP is connected (call get_session_identity).

Phase 1: Identify financial documents

1. Ask the user what to import:
   • Bank statement file path (CSV or PDF)
   • Receipt image(s)
   • Invoice file(s)
   • Live bank account via Plaid MCP
2. Detect the format and choose the appropriate parsing method.

Phase 2: Parse and preview

1. CSV statements: read columns, detect header row, identify date/amount/description fields.
2. PDF statements: use parse_file to extract text, then parse rows.
3. Receipt images: use vision to read merchant, date, items, total, tax.
4. Invoices: use parse_file to extract vendor, line items, amounts, due date.
5. Present a preview: transaction count, date range, total amounts, top merchants/vendors.
6. Ask for confirmation.

Phase 3: Extract entities

For each document, extract:

1. Transactions: one transaction entity per line item or statement row with amount, currency, date, description, merchant, category (when inferable).
2. Contacts: merchants, vendors, or counterparties as contact entities.
3. Invoices/Receipts: the document itself as a receipt or invoice entity with total, tax, line items.

Use entity_type transaction consistently within a batch. Store source-specific details as fields (provider, account_suffix, value_date, concept).

Phase 4: Store with provenance

Use the combined store path — entities array plus file_path for the original document — in a single store call. This preserves the raw financial document as a source row.

Set source_file to the filename. For each transaction, set data_source with a unique identifier (e.g. row number, transaction reference).

Phase 5: Report results

Summarize:

• Documents processed
• Transactions stored (count, date range, total)
• Top merchants/vendors by frequency or amount
• Offer to query specific transactions or merchants.

Do not

• Process financial documents without user confirmation.
• Mix entity_type within a batch (use transaction consistently for statement rows).
• Skip the raw document preservation — always use the combined store path.
• Echo full financial details in chat beyond what answering requires.