- Open the desktop app's file pickers in the folder of the last picked file, also after a restart
- Keep that folder in its own file in the app data folder, since the backend rewrites the global preferences file while it runs
- Fall back to the platform default when no folder is remembered or it no longer exists, and leave folder pickers on the default
- Add tests
Issue: actualbudget/actual#8887
* Fix linked account balances shown wrong or crashing for decimal amounts
* Show unknown instead of crashing for an invalid linked account balance
* [AI] Trim linked balance conversion regression tests
* [AI] Show unknown for linked balances too large to format
* [AI] Keep Enable Banking linked balances in cents
* [AI] Tighten linked balance helper comment and release note
---------
Co-authored-by: Tim Smart <hello@timsmart.co>
* [AI] Send large sync backlogs to the server in batches
Importing a budget with a long history into a fresh server makes the
client upload every CRDT message in one /sync request. A budget with
~185k messages produced a 38MB request, above the server's default 20MB
limit, so sync failed with PayloadTooLargeError.
_fullSync now splits the outgoing messages into batches of at most ~4MB
(estimated). The first batch asks for the server's messages since the
sync point, later batches only for what is newer than the local clock;
the merkle diff still catches anything missed in between.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FDDv6fYQmFRk6ebDsqc5Ux
* [AI] Reword release note in plain language
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FDDv6fYQmFRk6ebDsqc5Ux
* [AI] Measure sync batch sizes in UTF-8 bytes
String.length counts UTF-16 code units, which undercounts non-ASCII
text (accented letters, symbols, emoji) as it goes over the wire.
Count UTF-8 bytes instead so batches stay within the intended size.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FDDv6fYQmFRk6ebDsqc5Ux
* [AI] Simplify sync batching and cover later-batch responses
* [AI] Tidy sync batching comment and release note
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Tim Smart <hello@timsmart.co>
* [AI] fix: prevent background loader flicker
Keep the lazy importer stable across renders so LoadComponent does not unload and reload the background when its parent rerenders.
* docs: clarify background flicker release note
* test: assert background artwork identity synchronously
---------
Co-authored-by: Tim Smart <hello@timsmart.co>
* export the color picker palette
* add a color sheet for mobile
* add mobile tags list and edit pages
* route tags to the mobile pages and show them in mobile settings
* add unit tests for the mobile tags pages
* add e2e tests for the mobile tags pages
* note
* Update VRT screenshots
Auto-generated by VRT workflow
PR: #9509
* remove the edit page actions and use the default tag size in the list
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Matiss Janis Aboltins <matiss@mja.lv>
* [AI] Skip expired template targets instead of failing the run
A non-repeating `by` or `spend` target whose month had passed threw an
error, which stopped templates from budgeting any category that month.
Set expired targets aside so the rest of the run continues. Earlier
months can still be re-run, money left in the category isn't pulled
back, and a warning suggests removing a target once its month has
really passed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Show passed targets as a warning in the automations editor
The editor still treated a passed one-off target as an error that
disabled Save, even though the engine now skips it.
Move the expiry rule to a shared isTemplateExpired helper used by both
the engine and the editor. A passed target now shows as a warning
without blocking Save, and expired targets no longer count toward the
schedule and by-date priority check.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Shorten the passed-target warning so it fits the automation row
The row text was cut off. Use the same wording as the banner title.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* change release note category
* [AI] Never budget a negative amount for by and spend targets
The clamp only applied while a category still had an expired target, so
removing the expired line, as the warning suggests, let the next run pull
the leftover money back anyway. Clamp runBy and runSpend at 0 instead, so
money already set aside beyond a target always stays in the category.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Count expired targets in the editor's priority check
The editor skipped expired by targets when checking that schedule and by
priorities match, using today's month. Rerunning templates for the month
the target was active in still includes it, so the engine rejected the
mismatch and the whole run budgeted nothing. Count every by target again,
so anything the editor saves runs for any month.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Disallow negative amounts in by and spend templates
A negative target never budgets anything now that runBy and runSpend
clamp at 0. Reject it instead: notes templates with a negative by or
spend amount no longer parse, and the editor locks the amount to positive
and flags any saved before as a blocking error. Other template types can
still be negative.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Check spend errors before the passed-target warning
The passed-target warning was returned before the spend checks, so an
expired one-off spend target with a missing or later start month showed
only the warning and could be saved. Rerunning the month it was active
in then ignored early spending. Check the blocking spend errors first.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Express 5's app.listen passes its callback to server.once('error'), so a
failed bind (EADDRINUSE, EACCES) ran the success callback, printed
"Listening on ...", and the process exited 0 once nothing kept it alive.
Listen through a small helper that resolves on 'listening' and rejects on
'error', and return the startup promise from app.ts so the existing catch
logs the error and exits 1.
Assisted-by: ClaudeCode:claude-opus-5-5
* [AI] Keep the iOS App Store listing in the repo and publish it on release
Adds a fastlane metadata and screenshots folder for the iOS app and a
release-only workflow step that uploads them with fastlane deliver before
the build is submitted for review.
Fixes#9446
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* [AI] Add iPhone Duo App Store screenshots
Captured from the iPhone Duo simulator's unfolded display (2853x2007).
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* [AI] Keep the wide layout clear of the status bar in the mobile apps
The wide layout ignored safe-area insets, so on the unfolded iPhone Duo the
status cluster covered the title bar buttons. Pad the sidebar and content
column by the insets (zero on desktop and web) and recapture the Duo App
Store screenshots with the fix.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* [AI] Merge the two release notes into one
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
* [AI] Apply the left safe-area inset to the main pane when the sidebar floats
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
* feat(api): expose dashboard and widget management methods
* fix(api): address review comments on dashboard typing and atomicity
* test(api): cover multiple widgets in updateDashboard batch
* fix(dashboard): validate widget metadata on insertion and update
* fix(dashboard): sanitize add-widget coordinates and validate all updates
* fix(dashboard): preserve layout editing for invalid metadata, reject batch duplicates, and normalize imports
* test(dashboard): use unique temp file path in import test
* refactor(api): use send handlers for dashboard and remove runtime over-validation
- Add dashboard-get-pages and dashboard-get-widgets message handlers in loot-core
- Simplify getDashboardPages, getDashboardWidgets, and addDashboardWidget to thin send dispatchers in packages/api
- Remove runtime widget type assertions, metadata schemas, duplicate ID checks, and error throws on missing widgets in loot-core
- Revert importDashboard metadata normalization back to master
- Remove type coercions and validation rejection tests
- Update release notes category to Enhancements
* fix(api): address review comments on dashboard queries, typings, and documentation
---------
Co-authored-by: github-actions[bot] <matiss@mja.lv>
Co-authored-by: Claude Fable 5.2 <noreply@anthropic.com>
Add a dashboard-move handler that syncs the order of every dashboard.
You can drag dashboards in the dashboard selector, or move them with the keyboard.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* make payees, tags, rules, schedules and bank sync pages consistent
* title every page in the settings navigation
* drop the providers heading from the bank sync page
* spacing
* Update VRT screenshots
Auto-generated by VRT workflow
PR: #9218
* fix e2e
* coderabbit
* button colour
* Update VRT screenshots
Auto-generated by VRT workflow
PR: #9218
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
* [AI] fix: skip completed schedule rules when linking transactions
Completed schedules are kept for reference, but their rules still ran on any matching transaction and linked it to the schedule. runRules now skips the rules of completed, non-deleted schedules unless the transaction already belongs to a schedule. Existing links are untouched and reopening a schedule makes it link again.
Assisted-by: ClaudeCode:claude-sonnet-5-5
* [AI] refactor: dedupe getCompletedScheduleRuleIds and test schedule-linked runRules
Drop the AQL-based getCompletedScheduleRuleIds from schedules/app and
have payees use the one in transaction-rules, which schedules/app already
depends on. Replace the existing-links test with one that runs runRules
on a transaction linked to a completed schedule and checks the
schedule's rule still applies.
Assisted-by: ClaudeCode:claude-opus-5-5
* [AI] Manage budget API keys in settings
* [AI] Use modals for API key creation and revocation
* [AI] Simplify API key revocation confirmation
* [AI] Refine API key settings and modal interactions
* [AI] Remove trailing punctuation from API key release note
* [AI] Use budget grant lists in API key settings and documentation
* [AI] Focus API key UI tests on behaviour
* [AI] Warn that authentication changes revoke API keys
* [AI] Clarify which credentials password changes revoke
* Humanize docs
* Reduce
* [AI] Allow API keys to download and sync one budget
* [AI] Authorize API keys across their accessible budget grants
* [AI] Verify data isolation across API key budgets
* [AI] Simplify API key access regression tests
* [AI] Add budget API key lifecycle endpoints
* [AI] Clarify API key validation and preserve session boundary coverage
* [AI] Store API key grants for multiple budgets
* [AI] Remove API keys when their owner is deleted
* [AI] Revoke API keys when server authentication changes
* [AI] Preserve OpenID API keys when changing the server password
2026-11-08 19:12:21 +00:00
Michael ClarkGitHubautofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
* phase 2
* fix deploy
* [autofix.ci] apply automated fixes
* Trim the storage layer after review
* Keep the legacy database openable after a failed migration and release the pool on a failed open
* Copy the legacy database as one snapshot and share pool acquisition between concurrent opens
* Reject impossible page sizes and share one open per file
* Ask for OPFS whenever any budget has moved
* Share the budget lock from the legacy engine so migrations wait for it
* Release the legacy lock on a failed open, commit marker writes, keep unsaved snapshots and prefer the fallback row
* feedback
* bit of hardening
* nit
* feedback
* feedback
* feedback
* feedback
* feedback
* fix some minor issues
* minor nit
* cover removal and check marker survives
* directory removal now checks every db in teh tree before deleting everything
* batch removal
* copilot feedback
* minor nit
* add no op flushSqliteStorage for electron
* more copilot feedback
* legcy directory removal fix
* rabbit feedback
* feedback
---------
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
* [AI] Add hover button to remove a tag from a transaction's notes
Hovering a tag in the desktop transaction table's notes cell (or in the
truncated-notes tooltip) reveals a small remove badge in the pill's
top-right corner. Clicking it deletes that one tag occurrence from the
notes; clicking the rest of the pill still filters by the tag.
The badge is absolutely positioned so revealing it never shifts the
notes text. Removal works on source offsets recorded by the notes
parser, since parsed segments don't round-trip to the original string.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Update VRT screenshots
Auto-generated by VRT workflow
PR: #9093
* [AI] Drop unreachable stale-segment guard from removeTagFromNotes
Callers always pass a segment parsed from the same notes string, so the
check could never fire. Document the precondition instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DWkJZiCWJQ2sfzjbu5u9NR
* [AI] Annotate getTagCSSColors return type
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DWkJZiCWJQ2sfzjbu5u9NR
* [AI] Explain why only custom tag colours use !important
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DWkJZiCWJQ2sfzjbu5u9NR
* [AI] Clarify what a custom tag colour is in the !important comment
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DWkJZiCWJQ2sfzjbu5u9NR
* [AI] Rename useTagColors to useTagPillStyle
It also returns whether the tag is hidden, which isn't a colour.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DWkJZiCWJQ2sfzjbu5u9NR
* Update VRT screenshots
Auto-generated by VRT workflow
PR: #9093
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Categories in a hidden group still counted when "Show hidden categories" was
off, because hiding a group doesn't mark its categories as hidden. The report
now excludes them unless the option is on.
The first month's overspending adjustment was always 0, because the overspending
from the month before the period was dropped. The first month now shows it, and
the period total includes it. Balances are unchanged, since the adjustment is
informational.
Assisted-by: ClaudeCode:claude-opus-5-5
When advanceSchedulesService queries schedules to advance paid schedules, it filtered on '_account.closed': false. For schedules without an account, _account is null so _account.closed evaluates to NULL (failing the filter). Update the filter to allow _account: null in addition to open accounts.
Fixes#9463
* fix: keep line breaks in transaction notes when the notes field is only focused
* fix: stop NotesCell CustomCell from overwriting notes on blur
Remove the direct onUpdate prop from NotesCell's CustomCell so a stale
initial value cannot bypass the guarded onUpdateNote path. Add a test
that focusing a note with leading/trailing whitespace does not rewrite it.
* chore: retrigger CI
* [AI] Add API methods to budget a category from its templates
applyCategoryTemplate runs the same budget/apply-single-template handler as
the "Overwrite with template" item in a category's budget menu.
getCategoryTemplateAmount runs budget/dry-run-category-template, which the
budget automations editor uses for its preview. It reads the category's saved
templates rather than the ones open in the editor.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Throw from the template dry run when the templates fail validation
dryRunCategoryTemplate ignored the errors computeTemplates collects, so a
category whose templates fail validation, for example with two #goal lines,
was previewed as 0. It now throws with those errors. The automations editor
already catches a failed dry run. getCategoryTemplateAmount rethrows the
error as an APIError.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Return the template dry run's validation errors instead of throwing
Throwing from dryRunCategoryTemplate surfaced as an internal error in the
budget automations editor, for example when "Cover schedule" is picked
before a schedule is chosen. The dry run now returns the errors
computeTemplates collects alongside the amounts, and
getCategoryTemplateAmount throws an APIError when there are any.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Reject category templates that fail to parse in the template API methods
A #template line that fails to parse is stored as an error template, which
the template runner skips, so applying could budget 0 and report success.
applyCategoryTemplate and getCategoryTemplateAmount now throw with the
failing lines instead. The UI's apply paths are left unchanged.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Format the template API test
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Read category templates for the template API without storing them
The template preview parsed the category's note by storing its templates in goal_def first, so a read wrote to the budget. It now parses the note in memory, and applying stores the templates once, inside apply-single-template.
* [AI] Type the template API test's goal_def query
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* [AI] Send a plain IPv4 PSU address to Enable Banking and tolerate balance failures
Node reports an IPv4 client on a dual-stack socket as ::ffff:1.2.3.4,
and some banks (Zagrebačka banka, SEB) reject the PSU address in that
form, failing every sync. Convert it to plain IPv4 before sending.
A failed balances request aborted the sync before the transactions were
fetched, so banks whose balance endpoint fails never imported anything.
Carry on without a balance, as account discovery already does.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Report an unavailable Enable Banking balance as unknown instead of zero
When the bank refuses the balances request, the sync server now returns
no balance rather than 0. The client then imports the transactions
without inserting an opening balance on the first sync, and leaves the
stored account balance untouched on later ones.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Use strict equality in the Enable Banking opening balance checks
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Treat a missing Enable Banking balance field like a null one
Normalize an absent startingBalance to null where the server response
is read, so the strict null checks also cover a server that omits the
field.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Derive a transaction ID for Enable Banking transactions that have none
Some banks send neither entry_reference nor transaction_id, which left
imported_id empty. Such transactions could only be recognised by amount
and date, so a deleted one was imported again on every sync and the
"reimport deleted transactions" setting had no effect.
Derive an ID from the booking date, amount, direction, counterparty and
remittance text. Identical entries within one response get a counter
suffix so genuinely separate postings stay separate. Bank-provided IDs
are used unchanged.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Prefer the transaction date when deriving an Enable Banking transaction ID
A pending transaction has no booking date yet, so its derived ID was
built from a fallback date and changed once the transaction was booked
on a different day. Use the date of the transaction itself first, which
stays the same across that transition.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Number booked Enable Banking postings before identical pending ones
Identical postings without a bank ID are told apart by a counter. A
pending posting listed first took the first number, so once it was
booked elsewhere or disappeared the booked posting inherited its ID and
the sync updated the wrong transaction. Number booked postings first so
a pending one coming or going never renumbers them.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Derive Enable Banking IDs only for transactions that can be imported
A transaction dropped for an unusable date still took a counter number,
so an identical valid transaction changed ID once the dropped one was
gone. Derive IDs after normalization and only for importable
transactions.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Hash the booking date last when deriving an Enable Banking transaction ID
A pending transaction has no booking date, so hashing the booking date
whenever it exists gave a pending transaction without a transaction date
a new ID on booking even when its value date stayed the same. Use the
value date before the booking date.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Fix Enable Banking reauthorization and silent setup failures
Reauthorizing an expired Enable Banking account opened the link flow
without the account, and the account was never preselected because it
was still linked to an id the provider no longer returns. Pass the
account along and preselect it so its link is replaced in place.
Setup failures were reported as success: the setup modal looked for the
validation error one level too deep, and a failed bank list request
rendered an empty dropdown. Show Enable Banking's reason in both places
and trim the application id on the server.
Document the prerequisites, reauthorization, reverse proxy and
single-application limits for Enable Banking.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Keep the experimental release warning on the Enable Banking docs page
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Address review feedback on Enable Banking reauthorization
Show the reason when the bank list request itself fails, not only when
Enable Banking refuses it, under a neutral label.
Drop the replaced link when preselecting a reauthorized account, so
removing the preselected row unlinks the account instead of leaving it
on the expired id.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Keep the expired link when a reauthorized account's row is reassigned
Dropping the expired link at preselection meant that picking a different
account for the row silently unlinked the reauthorized account while the
button still read "Link accounts". Keep the link, and drop it only when
the user removes bank sync from that row.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Only preselect a reauthorized account for a bank account with the same name
Preselecting the reauthorized account for the sole returned bank account
could link a different account at the same bank into it. Require the
bank account to carry the name the account had when it was linked, and
leave the row for the user to choose otherwise.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Do not preselect an account when reauthorizing Enable Banking
Enable Banking issues new account ids on every authorization and Actual
stores no stable identifier for the bank account, so the reauthorized
account could only be preselected by position or by name. Either can
link a different bank account into an account that already has synced
history. Leave the row for the user to choose, as before, and describe
that flow in the docs.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Adjust the mobile transaction input to support copy/paste/select all functions
* [AI] Added more tests and release notes
* [AI] Hide the caret at first when first entering an initial amount to avoid it "spawning" mid-input/character
* [AI] treat pasted text as an amount and keep caret in place
* [AI] paste over part of the amount instead of replacing it
* [AI] Fix HTTP_PROXY support for SimpleFIN connections
* [AI] Isolate SSRF connection test from proxy environment
* [AI] Clear proxy URLs in direct SSRF test
* [AI] Adopt the UIScene lifecycle in the iOS app
Apps built with the iOS 27 SDK must use the scene-based lifecycle or they
crash on launch. Add a SceneDelegate that forwards to Capacitor's
SceneDelegateProxy and register it in Info.plist.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
* [AI] Update native-run so cap run ios works with Xcode 27
Xcode 27 replaced Simulator.app with DeviceHub.app; native-run 2.1.0
knows about it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>