What’s New: Bank Integrations, Binance Sync, Deposits, and Debt
A lot shipped between August and September 2026 with zero announcement. Here is what changed, grouped by area, with no marketing gloss — just what the app does differently today.

A lot shipped between August and September 2026 with zero announcement. Here is what changed, grouped by area, with no marketing gloss — just what the app does differently today.

Bank sync used to be a per-provider afterthought. It is now a capability-driven settings surface: each integration declares what it can do, the settings page renders those capabilities generically instead of hard-coding a screen per bank, and you can add more accounts from an already-connected bank without leaving the integration page. Home-screen accounts now group by integration instead of listing every synced account flat. See bank integration management for the full picture.
Binance accounts sync directly, including P2P trades reconciled against your bank transactions so a P2P buy and the matching bank transfer do not double-count as two separate expenses. Crypto grouping was also corrected twice this cycle: exchange-synced coins stay under their exchange section, while self-custodied coins group by currency instead — the two flows now stay visually and logically separate. Details on Binance sync.
Fixed-term deposits now have their own account type with an interest rate field, a routed creation form, and a Close Deposit action that transfers the balance to a destination account of your choice — including inline deposit creation directly from the convert-to-transfer picker. Deposits are excluded from expense-source lists by default, since maturing a deposit is not spending. See deposit tracking.
Debt balance semantics are now unified across the debt tile and the underlying ledger, the progress indicator was redesigned around a neutral module instead of a colored bar that implied good or bad, and settling a debt now categorizes the resulting expense as a dedicated Debt Payment instead of falling into a generic category. A backfill hardened existing debt settlements against the old categorization. See debt & loan tracking.
The transactions list now supports filtering by amount range in addition to date, category, tag, and account. Opening a transaction shows a read-only detail view with six-month similar-spend bars, so you can see whether an amount is typical for that merchant or category without leaving the screen. Several split-confirmation keyboard bugs were also fixed, so entering split amounts on a physical or software keyboard no longer misbehaves mid-edit.
Consolidation is what decides when two bank rows — a card payment and a settlement, a transfer leg and its counterpart — are really the same movement of money. This cycle tightened that engine on several fronts at once: Monobank payment-refund titles now match correctly, FX-tolerant bridge chains reclaim with a rebuild fallback instead of failing outright, absorbed transactions are restored rather than deleted when you unconsolidate, mutual-best refunds are accepted instead of dropping an over-sum group, a fee-return refund is recognized when the fee entry exceeds the primary credit, and a bridge family stands down cleanly when an existing same-pair canonical already covers it. None of this is visible in a screenshot — it shows up as fewer duplicate or orphaned transactions after a bank sync.
Running a large language model and a Whisper speech model on-device raises an obvious question: does the app stay fast to open? This cycle deferred llama and Whisper module evaluation past the first rendered frame, switched lucide icons to per-icon lazy imports, gated Lingui catalog loading behind an activation check instead of loading every locale eagerly, deferred sync services and background task registration off the boot path, dropped unreferenced embedded fonts, switched to date-fns deep subpath imports, and removed the intl-pluralrules polyfill now that Hermes ships it natively. Each change is small; together they keep first paint fast even with two on-device models available.
No. All of this ships in the regular app update. Deposit and debt accounts are opt-in — you only see the new fields when you create or edit that account type.
The accuracy fixes apply going forward and to the specific edge cases the backfills targeted. They do not rewrite unrelated transaction history.
This post groups the user-visible work from one release window. It is not a line-by-line release log — check the app store listing for the exact version history.
Join the Budgie waitlist and be the first to experience truly private expense tracking.
Join Waitlist2026-05-07
Cloud AI assistants process your spending data on remote servers. On-device AI keeps every inference local. Here's how a 1.7B-param model, offline embeddings, and Whisper speech recognition work together in Budgie.
Read article→2026-05-07
Financial aggregators centralise millions of bank credentials in one place — a magnet for attackers. Offline-first architecture eliminates the target. Here's how Budgie connects to banks without handing your credentials to a third party.
Read article→