Skip to main content
Didit’s Bank data API stores each connection’s accounts, balances and transactions and serves them through read endpoints with pagination (cursor on lists, page_token on the transaction reads). Amounts are decimal strings with a separate ISO 4217 currency, dates are ISO 8601, and countries are ISO alpha-3. Reads never bill.

Accounts

GET /v3/bank/connections/{id}/accounts/ lists the accounts of a connection, and GET /v3/bank/accounts/{account_id}/ returns one. Use account_ref to recognise a bank account when a person connects it again, and id for everything inside one connection.

Balances

GET /v3/bank/accounts/{account_id}/balances/ returns typed balances with their reference dates.
type is booked, available or expected, and iso_type carries the bank’s ISO 20022 code where the bank gives one.

Transactions

GET /v3/bank/accounts/{account_id}/transactions/ reads a date range. It takes date_from and date_to (calendar dates as the bank reports them, inclusive), status (booked or pending, both by default), limit (up to 500) and page_token. It sorts by effective_date descending, then id.
  • id is Didit’s, stored at first sight. The same movement keeps its id across reads, syncs and renewals.
  • identity_quality says how strong that guarantee is. bank_reference and route_id are stable. derived means the movement was matched by its content, so a bank correction can appear as one removed and one added.
  • status is booked or pending. A pending transaction has its own id. When the bank books it, the booked transaction arrives as added, the pending one as removed with reason: pending_booked, and pending_id is set only where the bank links them.
  • amount is signed: negative is money out. direction repeats the sign.
  • bank_reference is the bank’s immutable entry reference, null where the bank does not provide one. Didit never invents it.
  • category is Didit’s own categorisation, when one applies.
  • A correction by the bank to a booked transaction is modified. A reversal is a new transaction, with reverses_id when the bank links it.

Fields a bank does not send are null

Every field a bank does not send is null. The institution lists the fields it is known to send in transaction_fields, together with pending_transactions, pending_linkage, balance_types and history_days_max, so nobody discovers gaps in production. Use the sparse_fields test customer in the sandbox to test your code against a bank that sends little.

Paging a date range

Keep reading with page_token while has_more is true. The token pins the read’s snapshot for one hour: additions after the first page are not in it, each row is served in its current state, and a row removed during paging drops out of later pages and shows up in the change feed. An expired token answers 410 page_expired: restart the read from its original query.

The change feed

GET /v3/bank/connections/{id}/transactions/sync/ returns what changed across all accounts of a connection since a cursor.
  • An empty cursor returns every current transaction as added with reset: true. Replace your copy with it instead of adding to it.
  • Keep paging with page_token while has_more is true. Store next_cursor only after a response with has_more: false, and if paging fails midway restart from the cursor you stored.
  • A cursor is opaque and bound to the connection and the environment. The change log is replayable for 90 days. An older cursor answers 410 cursor_expired with bootstrap: true: start again from an empty cursor and replace your copy.
  • removed[].reason is pending_booked, pending_expired, bank_removed or access_removed.
The step by step version is in Continuous bank feed for accounting.

Headers

Both transaction reads return the X-RateLimit-* headers (600 reads and 300 writes per minute per application, 429 with Retry-After beyond that) and X-Bank-Last-Synced-At. API limits and the bank’s own limits on fetching are separate things: see Refresh bank data on demand.
Last modified on November 22, 2026