* fix(tinkoff): resolve the tradeable listing and price bonds via BondBy
Validated against the live T-Invest API and fixed three correctness issues in
Provider::TinkoffInvest:
- FindInstrument returns several listings per ticker (e.g. SBER on TQBR plus
dark/non-API boards 37M/SPEQ); only the apiTradeAvailableFlag listing has live
prices. resolve_short now ranks tradeable-first (then requested-MIC, then exact
ticker/ISIN) instead of taking the first match, so GetLastPrices/GetCandles
return data. Search now surfaces only tradeable instruments.
- find_instruments no longer hard-filters on apiTradeAvailableFlag, so a
qualified-investor instrument Tinkoff lists but can't API-trade still resolves
(for logos) and otherwise falls back to another price provider.
- Bond nominal comes from BondBy (the generic GetInstrumentBy omits it),
returning the current amortized nominal so percent-of-par converts correctly.
Verified end-to-end: SBER 313.65, T 282.76, LQDT 2.019, and the SFO Split bond
(amortized nominal 417.71) all price; logos resolve to real CDN PNGs.
* fix(tinkoff): honor requested MIC first, guard amortizing bonds, strip suffix
Address review feedback on the resolution/pricing path:
- resolve_short now ranks a requested-MIC match BEFORE tradeability, so a
security is never priced off another exchange's listing (e.g. a MISX security
no longer picks up a tradeable XSPX board's price); tradeability and exact
ticker/ISIN remain secondary tie-breaks.
- Amortizing bonds: BondBy returns only the current nominal, so applying it to
historical percent-of-par closes would underprice them. For amortizing bonds
we now return only the live price and skip the candle history; fixed-par bonds
and equities keep full history. bond_info exposes nominal + amortizationFlag.
- Strip exchange suffixes (.ME/.MOEX/.MISX/.MCX) before querying T-Invest, which
only knows the bare SECID — so a stored "T.MOEX" ticker resolves to "T".
* fix(tinkoff): don't cache nil resolution results (skip_nil) so a transient empty response isn't locked in for the 24h TTL
* feat(mobile): add SureCard primitive and migrate account cards
SureCard mirrors the web card chrome — `bg-container` + a hairline border +
rounded corners + the subtle DS shadow (`shadowXs`, from the scale shipped in
#2349) — resolved from the active SureColors palette so it stays in lockstep with
`sure.tokens.json` in light and dark. An optional `onTap` gives a flat ink
response clipped to the card radius.
Migrates AccountCard off the Material `Card` to `SureCard`.
Refs #2235.
* fix(coinstats): deterministic wallet batch order in bulk fetches
The bulk balance/transaction fetches built the "blockchain:address" param in
linked-account query order, which isn't stable across runs — so the batched
param (and its mocked expectations) varied seed to seed, intermittently failing
CoinstatsItem::ImporterTest and red-flagging unrelated PRs. Sort the wallets
before joining so the batch order is deterministic, and align the test
expectations to the sorted order.
* fix(mobile): tokenize swipe-reveal radius + cover SureCard dark theme
Address PR review:
- The account-card swipe-reveal background still used a hardcoded radius (12)
that mismatched SureCard's radiusLg (10) once the card was migrated — tokenize
it so the reveal corners line up during the swipe.
- Parameterize the SureCard chrome test over light + dark, since the card is
brightness-aware (catches palette regressions in either mode).
* test(mobile): assert SureCard onTap ink is clipped to the card radius
Address review nit: the onTap test claimed the ink is clipped to the card but
only checked the InkWell exists. Now it asserts the InkWell's borderRadius equals
SureTokens.radiusLg (tester.widget already enforces a single InkWell).
* feat(up): add Up Bank (AU) provider integration
Adds Up Bank as a per-family, token-based bank sync provider, modelled on
the existing Akahu integration. Up uses a JSON:API REST API with a personal
access token (Bearer), cursor pagination via links.next, and returns both
HELD (pending) and SETTLED transactions from one endpoint.
New:
- Provider::Up client (JSON:API unwrap, links.next pagination, retries,
typed errors, /util/ping) + Provider::UpAdapter (Factory-registered,
Depository + Loan).
- UpItem / UpAccount models with Provided, Unlinking, Syncer,
SyncCompleteEvent, Importer, Processor, Transactions::Processor, and
UpEntry::Processor (amount sign flip, HELD->pending, foreignAmount FX,
merchant from description, stale-pending pruning).
- Family::UpConnectable, UpItemsController, routes, settings panel + connect
flow views, accounts index wiring, initializer, en locale, and model tests.
Core wiring:
- "up" added to Transaction::PENDING_PROVIDERS, the three pending-match SQL
blocks in Account::ProviderImportAdapter, Provider::Metadata::REGISTRY,
ProviderMerchant/DataEnrichment source enums, ProviderConnectionStatus,
settings provider panels, and financial data reset.
Migration create_up_items_and_accounts must be run before use. No external
API endpoints added (no OpenAPI changes).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(up): dump up tables to schema and make since filter TZ-safe
The feature commit added the up_items/up_accounts migration but never
re-dumped db/schema.rb, leaving the schema version and tables stale.
Add the two table definitions and foreign keys and bump the schema
version so a fresh DB load matches the migration.
Also format a bare Date `since` as UTC midnight instead of the server's
local zone, so `filter[since]` is deterministic regardless of where the
app runs (previously shifted by the local UTC offset).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(up): address code review feedback
Behavior/correctness:
- Persist skipped accounts via a new up_accounts.ignored flag and a
needs_setup scope, so skipped accounts stop resurfacing as "needs
setup" on every sync. Linking clears the flag.
- destroy now checks unlink_all! per-account results and aborts deletion
(alert) if any unlink failed, instead of swallowing failures.
- render_provider_panel_error redirect uses :see_other (was an invalid
4xx redirect status).
- Up provider adapter falls back to item institution name/url when
institution_metadata is absent (early return previously blocked it).
Resilience/security:
- fetch_all_resources guards against an API repeating the same
links.next cursor (Set#add?), preventing infinite pagination.
- HTTP client validates absolute URLs (from links.next) against Up's
HTTPS host before sending the bearer token, preventing credential
leakage to untrusted hosts.
Diagnostics:
- Route provider sync/import failures through DebugLogEntry.capture
(controller, UpItem, syncer, unlinking) with family/account context.
Low-level HTTP client and currency-normalization warnings keep
Rails.logger to match existing provider conventions.
Data integrity:
- up_accounts.name and currency are NOT NULL (align with model presence
validations); account_id stays nullable (allow_nil uniqueness).
Forms:
- select_existing_account radio is required; controller guards a blank/
unknown up_account_id with a friendly alert instead of RecordNotFound.
Tests:
- Add UpAccount needs_setup scope test, pagination loop guard test,
untrusted-host rejection test; tighten filter[since] assertion to the
exact UTC timestamp.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(up): address second-round review feedback
- Capture sync/import failures via DebugLogEntry so swallowed errors in
account/transaction fetching and transaction processing surface in
/settings/debug instead of only Rails.logger.
- Gate UP_DEBUG_RAW raw payload dump to local envs to avoid leaking PII
(merchant names, amounts, account IDs) in managed/production logs.
- Collapse linked/unlinked/total account counts into one memoized query
instead of 3 separate COUNTs per rendered item.
- Rename "Set Up Up Accounts" locale title to "Link Up Accounts".
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* docs(up): add method docstrings and align failed_result keys
Add docstrings to all Up provider source files (controller, models,
providers, concerns) to satisfy the 80% docstring coverage threshold.
Third-round review: failed_result now mirrors import's result shape
(accounts_updated/created/failed, transactions_imported/failed) instead
of the stale accounts_imported key, so failure results stay consistent.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Build fresh API session contexts instead of reusing persisted web sessions that may carry impersonation state.
Reject deactivated report export API key owners and strengthen regression coverage for API key, OAuth, and report export authentication paths.
Adds Provider::TinkoffInvest, a token-based securities provider built on the
public T-Invest REST gateway (invest-public-api.tinkoff.ru/rest). It serves
prices for Russian instruments (shares, ETF/БПИФ, bonds) and, crucially, brand
logos via the T-Invest CDN — the authoritative logo source for MOEX
instruments, which ISS (MoexPublic) does not provide.
- Registry: register `tinkoff_invest` under the :securities concept; token via
ENV TINKOFF_INVEST_API_KEY or encrypted Setting.tinkoff_invest_api_key.
- Logos independent of the price provider: Security#import_brand_logo consults
T-Invest for a logo whenever a token is configured (after the price-provider
metadata fetch, so it never short-circuits website_url backfill). Gated on
token presence, not the securities checklist.
- display_logo_url: with no website domain, a stored provider logo (T-Invest)
now beats the ticker-only Brandfetch lettermark; when a domain exists,
Brandfetch still wins (unchanged).
- MoexPublic no longer reports moex.com as the issuer website — it's the
exchange, not the issuer, and would make Brandfetch render the exchange logo
for every instrument and shadow the real brand logo.
- Prices: GetCandles (daily, paged) + GetLastPrices; Quotation units+nano/1e9;
bonds priced as percent-of-par x nominal (missing nominal raises, not 0).
- Settings: encrypted token field (always shown) + provider checkbox + en locale.
- Tests for search/info/logo-url/prices/bond/incomplete-candle and display logic.
Co-authored-by: Claude <noreply@anthropic.com>
* perf(reports): avoid residual category lazy loads
* test(reports): reuse SQL query capture helper
Move the reports SQL capture regression helper into test/support and document the category loading and budget reload choices called out in review.
* perf(dashboard): streamline investment activity totals
* fix(dashboard): skip empty investment totals cache writes
Short-circuit empty investment account totals before cache fetch and move the new SQL query capture regression helper into test/support.
* refactor(dashboard): address review on investment totals query
- Drop defensive ArgumentError guards in InvestmentStatement::Totals#initialize.
The sole caller (totals_query) always passes correct types, and the
empty_result early-return already handles the zero-accounts case.
- Remove the redundant accounts JOIN and its family_id/status filters from the
aggregation SQL. account_ids is already scoped to the family's visible
(draft/active) investment accounts, so the join back to accounts is
unnecessary work in the query plan. Drop the now-unused family_id param.
- Add a regression assertion that the aggregate no longer joins accounts.
* feat(prices): add Moscow Exchange (MOEX ISS) securities + FX provider
Add Provider::MoexPublic, a keyless provider built on the free MOEX ISS API
(https://iss.moex.com/iss), modeled on Provider::BinancePublic.
Securities: shares, funds/ETF/БПИФ (e.g. LQDT), and bonds (OFZ + corporate).
Bonds are priced clean — LAST% × FACEVALUE / 100 in the instrument currency,
with per-row FACEVALUE for amortizing issues; NKD/accrued coupon excluded.
Exchange rates: also implements ExchangeRateConcept for RUB↔{USD,EUR,CNY} via
selt TOM instruments (USD000UTSTOM/EUR_RUB__TOM/CNYRUB_TOM); the selt quote is
X/RUB, inverted for RUB→X, nil for non-RUB-crossed pairs.
Details:
- Board/engine resolution via the ISS primary-board flag with a hardcoded
priority fallback (TQBR, TQTF, TQOB, TQCB, …).
- Instrument currency from CURRENCYID/FACEUNIT (handles USD/CNY eurobonds &
FX funds), normalizing legacy SUR/RUR → RUB; default RUB.
- Full history via from/till + start= pagination; current price fallback chain
LAST → MARKETPRICE → LCURRENTPRICE → LCLOSEPRICE → PREVPRICE → latest history
close.
- Bare SECID identity, exchange_operating_mic=MISX, country_code=nil (wildcard
like Binance); search accepts .ME/.MOEX/.MISX/.MCX aliases and ISIN.
- RateLimitable throttling, SslConfigurable, Faraday retry/timeouts; all public
methods wrapped in with_provider_response.
Wired into Provider::Registry for both :securities and :exchange_rates, the
hosting provider-selection UI, locales, and config/exchanges.yml (MISX).
Docker-tested (devcontainer, Ruby 3.4.9): 29 new tests green, full provider
suite + i18n green, rubocop clean; smoke-tested against live ISS (SBER price,
OFZ clean price, USD/RUB FX).
* fix(moex): address review — FX weekend lookback, dead branch, translated hints
- fetch_exchange_rate now fetches a 10-day lookback window (not just the exact
day) so a weekend/holiday request resolves to the prior trading day's close,
matching Yahoo's behavior (Codex P2).
- Remove dead identical if/else branches in history_row_price (CodeRabbit).
- Translate moex_public_hint into ca/fr/hu/vi/zh-CN instead of English copy
(CodeRabbit).
- Add a test covering the FX prior-trading-day lookback.
* fix(moex): guard ISS date parsing; doc TQTE in board priority
Address maintainer review (jjmata):
- parse_iss_date wraps Date.parse so a malformed ISS TRADEDATE skips just that
row (with a contextual log warning) instead of failing the whole history/FX
fetch. Used in history_row_price and fx_history.
- Add TQTE to the BOARD_PRIORITY doc comment (it was in the constant but missing
from the comment).
- Add a test covering the unparseable-date skip.
* fix(prices): resolve dashed crypto tickers (BTC-USD, TRX-USD) via BinancePublic
Provider::BinancePublic is the dedicated crypto price provider, but its
search_securities only matched bare base assets ("BTC") or unseparated
Binance pairs ("BTCUSDT"/"BTCUSD"). The canonical "<BASE>-USD" form that
Yahoo emits and users paste (e.g. "TRX-USD", "USDT-USD") never matched, so
those holdings fell through to an unpriced offline security whenever the
stock provider didn't also return the coin.
Collapse the base/quote separator in the search query (and strip it in
parse_ticker) so "BTC-USD" / "TRX/USDT" are treated like "BTCUSD" and
resolve to live Binance pricing. Stablecoins pasted as "USDT-USD" resolve
to the synthetic USD price via their stripped base.
* fix(prices): only synthesize stablecoin search for USD quote
Address review: gate the synthetic stablecoin result on a USD quote (bare
coin or "<coin>USD" form). A non-USD quote like "USDTEUR" / "USDT-EUR"
has a real Binance pair and now falls through to normal matching instead of
being replaced by the USD synthetic. Drops the now-unused base_asset_query
helper. Adds a USDT-EUR regression test.
Import::UploadsController#update persists CSV uploads with
save!(validate: false), which skips the before_validation
:ensure_utf8_encoding callback. Non-UTF-8 files (e.g. ISO-8859-1 /
Windows-1252 exports from Brazilian banks) were written to the text
column as-is and rejected by Postgres with PG::CharacterNotInRepertoire,
surfacing as a 500 during upload.
Also register ensure_utf8_encoding on before_save so the existing
normalization (rchardet detection + Latin-1/Windows fallback) still runs
when validations are skipped. The callback is idempotent and no-ops on
valid UTF-8, so the validated path is unchanged.
Fixes#2294
The Account#subtype= writer delegates to accountable&.subtype=, which is a
silent no-op while the accountable is nil. On create the accountable is built
from accountable_attributes via accepts_nested_attributes_for, but the form
submits subtype as a top-level account attribute. Mass-assignment applies
subtype before accountable_attributes, so the selected subtype was dropped on
create (update worked because the accountable already exists).
Build the accountable from the delegated type inside the writer when it is not
yet present, so the value is preserved; the later accountable_attributes
assignment (update_only) updates the same record.
Add regression tests covering the create flow and the assignment ordering.
Adds YnabImport (mirroring ActualImport) for YNAB "Export budget" register CSVs:
- Amount — combines the split Outflow/Inflow columns into a single signed amount
(inflow - |outflow|), stripping currency symbols and thousands separators. A
single signed Amount column takes precedence when present.
- Category — resolves across export shapes: the combined "Category Group/Category"
column, the split "Category Group" + "Category", or legacy YNAB 4
"Master Category" + "Sub Category".
- Names — falls back from a blank Payee to the Memo, then the default row name.
- Validation — requires at least one amount source (Outflow/Inflow or Amount); a
file exposing none leaves rows un-clean instead of importing zero-dollar entries.
Enables the previously-disabled YNAB option on the imports screen (using the YNAB
logo, like Mint) with its configuration partial, and removes the now-dead
imports.new.coming_soon locale key. Documents the type in the API import-type
enums (rswag request spec + swagger_helper + generated openapi.yaml).
Closes#1255.
* feat(security): warn when ActiveRecord encryption is not configured
Self-hosted instances without explicit ACTIVE_RECORD_ENCRYPTION_* keys (or Rails credentials) store sensitive columns - API keys, provider/bank tokens, the MFA (TOTP) secret, and PII - unencrypted at rest. The app boots and works normally so this plaintext at rest state is easy to miss.
Change: Make it visible:
- log a clear startup warning (config/initializers/encryption_warning.rb)
- show a warning banner on /settings/security when encryption is unconfigured
* refactor(security): apply review feedback on encryption warning
- list the three ACTIVE_RECORD_ENCRYPTION_* keys in the banner, rendered via the DS::Alert content block to match the log
- drop the redundant respond_to?(:self_hosted?) guard in the initializer so it matches the controller check
- add a managed-mode test asserting the banner is hidden
When a rule is re-applied from the UI, RulesController passes
ignore_attribute_locks: true, but Enrichable#enrich_attributes still
rejected locked attributes unconditionally, so locked (manually edited
or import-locked) transactions were silently skipped and reported as
blocked.
Thread the flag through enrich_attribute/enrich_attributes as a new
ignore_locks keyword (default false, so provider syncs and AI
enrichment keep respecting locks) and pass it from the six synchronous
rule action executors.
Fixes#2051
* fix(jobs): enqueue jobs after transaction commit to fix SyncJob deserialization race
Syncable#sync_later creates a Sync row and enqueues SyncJob inside the
same transaction. Rails 7.2 deferred such enqueues until commit by
default; Rails 8.0 changed the default to enqueue immediately and 8.1
left the global config toggle non-functional, so a worker could dequeue
the job before COMMIT, fail to resolve the Sync GlobalID
(ActiveJob::DeserializationError), and have it silently dropped by
discard_on -- surfacing as stuck syncs after the Rails 8.1 upgrade.
Set enqueue_after_transaction_commit = true on ApplicationJob (the
Rails 8.2 default, inherited by every job) and drop the dead :never
symbol override on DestroyJob. Add regression and invariant tests.
* test(jobs): use OpenStruct for DestroyJob failure case
Replace the bare mock and the respond_to? expectation (an implementation
detail of how DestroyJob probes the model) with an OpenStruct that genuinely
responds to scheduled_for_deletion. Keeps only the command-facing assertions:
destroy raises and update! is called with scheduled_for_deletion: false.
Matches the repo convention of preferring OpenStruct for mock instances.
* test(snaptrade): assert connection-cleanup enqueue defers until commit
SnaptradeAccount#after_destroy enqueues SnaptradeConnectionCleanupJob, which
references the item/account by id. Before the ApplicationJob fix, Rails 8.1
enqueued it immediately inside the destroy transaction, so a worker could run
before COMMIT, see the not-yet-deleted row in its shared-authorization guard,
skip the provider call, and leak the SnapTrade connection. This regression
test destroys an account inside a transaction and asserts the job is not
enqueued until the transaction commits.
* feat(dashboard): masonry packing + per-widget size controls
In two-column mode the dashboard used a row-based CSS grid, so cards
stretched to equal row height and left dead space (e.g. the Net Worth
chart padded out to match the tall Balance Sheet table). Replace the
row-based layout with masonry packing and add per-widget size guardrails.
- Masonry: CSS grid with computed row-spans + grid-auto-flow: dense, driven
by a dashboard-masonry Stimulus controller (ResizeObserver + turbo:frame-load).
The DOM stays a single flat list, so drag/keyboard reorder is unaffected.
Active only in multi-column mode; single column falls back to normal flow.
- Internal sizing: the net worth chart height is now driven by a
--dash-widget-h CSS var (fixes an inert flex-1) so the card no longer pads
out below the chart.
- Guardrails: per-widget layout metadata (col_span, grow, min_height,
width_toggle) in PagesController, with per-user overrides persisted under
preferences["dashboard_section_layout"], deep-merged so width and height coexist.
- Size menu: a hover control on size-capable cards — Width (Half/Full) for the
cashflow sankey and net worth chart, Height (Compact/Auto/Tall) for grow
widgets. The sankey defaults to full width.
Adds model + controller tests for preference persistence and i18n keys.
* refactor(dashboard): redesign size menu with segmented controls
The size menu used a plain radio list and a diagonal maximize-2 trigger that
collided with the cashflow sankey's modal-expand button. Replace it with a
layout-config popover: a sliders-horizontal trigger plus two labeled axis
groups (Width, Height), each rendered as a DS::SegmentedControl with the
active option filled. Clearer, more compact, and it reads as a card-layout
control. The widget-size controller now mirrors the segmented control's
active-class + aria-pressed contract.
* feat(dashboard): expose width toggle on balance sheet and investments
Tables benefit from horizontal room, so give Balance Sheet and Investments
the same Width (Half/Full) control as the sankey and net worth chart. Height
presets stay chart-only — a table sized to a fixed height would just add
whitespace or force scrolling. The Outflows donut is intentionally left out
(full width is mostly whitespace for a donut).
* fix(dashboard): address review feedback on size controls
- Only apply the full-width col-span and show the Width control when the
two-column layout is enabled. A full widget previously leaked
2xl:col-span-2 into the single-column grid, creating an implicit second
column at 2xl widths and breaking the single-column preference. This also
keeps the Width control coherent with the Appearance two-column setting
(it now appears only where it does something). [Codex]
- Stop size-menu keydowns from bubbling to the section reorder handler, so
keyboard users can open the menu and pick options without entering
grab/reorder mode. [Codex]
- Harden preferences params: ignore a malformed (non-hash)
dashboard_section_layout / collapsed_sections instead of raising a 500,
and require section_order to be an array. [CodeRabbit]
- Localize the dashboard sections aria-label. [CodeRabbit]
* chore(settings): mention per-widget size controls in two-column copy
Surface the new per-widget width/height controls in the Appearance
"Two-column layout" description so the capability is discoverable.
* test(dashboard): assert non-mutation for malformed layout input
Addresses review feedback: asserting assert_nil made the test depend on
the fixture happening to have no dashboard height for net_worth_chart.
Capture the pre-PATCH value and assert it is unchanged, so the test
stays valid (malformed input ignored) even if the fixture later gets a
default height.
---------
Signed-off-by: Juan José Mata <juanjo.mata@gmail.com>
Co-authored-by: Guillem Arias <guillem.arias@col.vueling.com>
Co-authored-by: Juan José Mata <juanjo.mata@gmail.com>
* test: cover CoinStats missing-wallet balance preservation
* fix: preserve CoinStats balances when wallet data is missing
* chore: route CoinStats sync warnings to debug log
* docs: make debug log guidance timeless
---------
Signed-off-by: Juan José Mata <juanjo.mata@gmail.com>
Co-authored-by: Juan José Mata <juanjo.mata@gmail.com>
Commit 5d0eb7f4 replaced the color picker's DS::Disclosure with a raw
<details> because DS::Disclosure wraps its body in an mt-2 div, and that
normal-flow margin shoved the form down ~8px whenever the absolutely-
positioned picker popover opened. DS Drift Patrol flagged the raw
<details> in #2272 and #2316.
Add a body_class: option to DS::Disclosure (default mt-2) so callers can
drop the body margin, and migrate the color picker back onto the
component with body_class: nil. The summary's accessible name moves from
aria-label to an sr-only span in the summary content (verified: the
summary still reads as "Choose color and icon").
* fix(chat): clear the assistant message bubble when a turn is destroyed
When an assistant turn fails before any text streams (e.g. a provider
auth/model/network error on the first call), Assistant::Builtin#respond_to
destroys the still-pending message. Message only broadcast on create and
update, never on destroy, so the rendered 'Thinking…' bubble was never
removed — the chat appeared stuck thinking forever even though the job
had already errored (and appended an error via chat#add_error below it).
Add after_destroy_commit broadcast_remove_to so a destroyed message is
removed from the page.
* refactor(chat): trim destroy-broadcast comment to one line
Project convention asks for comments only when the why is non-obvious; the
behaviour is already covered by the commit/PR description. Per review feedback.
---------
Co-authored-by: Guillem Arias <guillem.arias@col.vueling.com>
* chore(deps): upgrade Rails 7.2 → 8.1
Rails 7.2 reaches end of life on 2026-08-09. Bump the framework to the
current 8.1.x line.
- Gemfile: rails "~> 8.0" (resolves 8.1.3); bundle update rails pulls the
Rails 8 framework gems plus the bumps it requires — ViewComponent
3.23 → 4.x (Rails 8 support), rails-i18n 7 → 8, rswag, and transitive deps.
- app/models/transfer.rb: make Transfer#date nil-safe
(inflow_transaction&.entry&.date). Rails 8's date_field evaluates the
field default on a new/unpersisted Transfer (the new-transfer form), where
the association is nil; without this, TransfersController#new raises
"undefined method 'entry' for nil". Matches the &. pattern already used in
Transfer#sync_account_later.
Framework behavioral defaults are unchanged (config.load_defaults stays as-is).
Validated on Rails 8.1.3: zeitwerk:check passes, full suite green
(4904 runs, 0 failures, 0 errors), rubocop and brakeman clean.
* fix(rails8): style textarea + deterministic property edit system test
The Rails 8 gem bump kept config.load_defaults at 7.2, but Rails 8 renamed
two ActionView::Helpers::FormBuilder field helpers regardless of defaults:
:text_area → :textarea and :check_box → :checkbox. StyledFormBuilder builds
its styled helpers from `field_helpers`, so `form.text_area` (e.g. the
account "Notes" field) silently fell through to the unstyled base helper and
rendered without a label — failing 8 system tests with
`Unable to find field "Notes"`.
- app/helpers/styled_form_builder.rb: exclude both spellings of the
non-text helpers (:check_box and :checkbox) and alias the legacy
`text_area` to the Rails 8 `textarea` so existing call sites stay styled.
Harmless on Rails 7.2 (old names present instead).
- test/system/property_test.rb: open the property edit dialog via the
account menu with a retry. The account page issues a Turbo morph refresh
shortly after load (turbo_refreshes_with :morph + a family-stream
broadcast); opening the modal while that refresh is in flight let the
morph re-render the page and wipe the just-loaded #modal turbo-frame.
Rails 8 timing made the race deterministic. Retrying once the refresh has
settled makes the test stable (confirmed via Turbo frame-load vs
full-page morph event traces; 3x green in isolation).
- config/brakeman.ignore: the added comment block shifted the pre-existing
(already-ignored, Weak) class_eval Dangerous Eval warning from line 5 -> 10,
changing its fingerprint. Re-point the existing suppression to the new
fingerprint/line so scan_ruby stays green.
Validated on Rails 8.1.3: full system suite green
(92 runs, 355 assertions, 0 failures, 0 errors), rubocop clean,
brakeman 0 warnings, CodeRabbit no findings.
* chore(deps): pin rails to the 8.1 minor line (~> 8.1.0)
Tighten the constraint from `~> 8.0` to `~> 8.1.0` (>= 8.1.0, < 8.2) so a
future `bundle update rails` tracks the 8.1.x line rather than silently
jumping to 8.2 when it ships. Matches the upgrade plan's stated intent
(target 8.1.x for the EOL runway) and a review note on #2301.
No resolved-version changes: bundle install keeps rails at 8.1.3 and every
other locked gem unchanged — only the Gemfile.lock DEPENDENCIES constraint
line moves. zeitwerk:check still passes; the already-green unit/system
suites ran on this exact resolved tree.
* chore(rails8): adopt Rails 8.1 framework defaults (config.load_defaults 8.1)
The gem bump above kept config.load_defaults at 7.2 so the change set could be
reasoned about in stages; this finalizes the upgrade by adopting the modern
framework defaults now that the suite is green on Rails 8.1.
Rails 8.0 added no new framework defaults (there is no new_framework_defaults_8_0
template), so 7.2 -> 8.1 is the single meaningful step. No incremental
new_framework_defaults_8_1.rb opt-in file is needed: the full suites pass with all
8.1 defaults enabled at once.
The 8.1 defaults this turns on include action_on_path_relative_redirect=:raise
(open-redirect hardening), raise_on_missing_required_finder_order_columns,
escape_json_responses=false / escape_js_separators_in_json=false (JSON perf), and
Ruby-parser template-dependency tracking.
Validated with no application code changes: bin/rails test 4904/0/0,
bin/rails test:system 92/0/0, rubocop + brakeman clean.
* chore(ci): restore brakeman CheckEOLRails now that the app is on Rails 8.1
config/brakeman.yml existed only to skip brakeman's CheckEOLRails. That check
fires on the calendar (it warns 60 days before a framework's EOL and escalates
as the date nears), so Rails 7.2's 2026-08-09 EOL turned `bin/brakeman` red
(exit 3) on every branch and on main regardless of the diff. The skip carried a
TODO to remove it once Sure upgraded off 7.2.
This PR puts the app on Rails 8.1 (EOL well in the future), so the skip is
obsolete; remove the file (its sole content was the skip) in the same change that
makes it unnecessary -- no stale-config window. brakeman auto-loads the file when
present and falls back to defaults when absent, and nothing references it
explicitly. CheckEOLRuby was already enabled and is unchanged; config/brakeman.ignore
is untouched.
Validated on Rails 8.1: bin/brakeman runs EOLRails + EOLRuby, 0 warnings,
0 errors, exit 0.
* feat(mobile): standardize money typography and semantic amount color
Add a brightness-aware SureColors theme extension and a MoneyText/SureMoney
primitive (semantic success/destructive/subdued tokens + tabular figures for
column-aligned digits), then migrate the transaction lists and balance cards
off raw Colors.green/red/grey.
Step 2 of the mobile design-system sequence (#2235), after #2237's theme
foundation. Primitive-first: screens consume shared tokens/typography.
* fix: review feedback — brightness-aware token fallback, de-flake Setting tests, Pipelock localhost FP
- SureColors.of falls back to the palette matching the active brightness (not
always light) when the extension is missing, so dark surfaces stay correct.
- Clear the rails-settings-cached cache before each test; its in-memory cache
survives the per-test transaction rollback, leaking Setting.* across tests and
flaking Settings::HostingsControllerTest (stale empty string vs nil).
Full unit suite: 4952 runs, 0 failures.
- Suppress the localhost test-DB DATABASE_URL false positive with line-level
`# pipelock:ignore` in ci.yml + llm-evals.yml instead of excluding whole files,
so those workflows stay scanned for real secrets.
* Update trades api with support for additional types
* rubocop fixes
* fix missing amount validation for interest type
* define missing schema reference
* fix test api_headers to use display_key per guidelines
* expand test coverage
* replaced duplicate JSON response blocks with helper method
* Add DB assertions to linked transfer test and fix invalid date test
* update brakeman.ignore fingerpint for refactored code
* Update the Brakeman ignore note to document validation for newly permitted keys
* fix API key auth in Minitest test to follow correct pattern
* update required in trades rswag spec to match the minimum fields that apply to all types
* extract dividend handling from build_investment_trade_params to dedicated method
* adjust response format to use the existing jbuilder views for Transfers and Transactions
* normalize type before passing to create form
* validate amount as a positive numeric value + tests
* rubocop fixes
* Add missing Trades API test coverage and docs
- Add Minitest tests for withdrawal (422, transfer linking), interest
(explicit ticker), and dividend update
- Add rswag 401/403/404/422 response docs for create, update, destroy
- Regenerate docs/api/openapi.yaml
* Update Security.find line reference in brakeman.ignore note
* Mark TransactionResponse account_type as nullable in rswag docs
* feat(cashflow): deep-link category labels to filtered transactions
Clicking a category's text label in the dashboard cashflow Sankey chart
now navigates to the transactions page filtered by that category and the
cashflow's active period date range. The colored node bar keeps its
existing zoom-into-subcategories behavior; structural nodes (Cash Flow,
Surplus) do not navigate.
URL-building lives in a pure, unit-tested utils/transactions_filter_url
module (mirroring utils/sankey_zoom) pinned in the importmap. Period dates
are threaded from the dashboard view into the Stimulus controller via data
values. This matches the existing donut chart's click-to-filter behavior.
* Update app/javascript/controllers/sankey_chart_controller.js
Co-authored-by: Guillem Arias Fauste <gariasf@proton.me>
Signed-off-by: Will Wilson <will@willwilson.uk>
---------
Signed-off-by: Will Wilson <will@willwilson.uk>
Co-authored-by: Guillem Arias Fauste <gariasf@proton.me>
PlaidAccount::TypeMappable maps the Plaid loan subtypes "home equity",
"line of credit", and "business" to home_equity, line_of_credit, and
business — but Loan::SUBTYPES never defined them. Linking any such
account (e.g. a HELOC reported by the institution as loan/"line of
credit") makes the item's sync fail with:
Validation failed: Accountable subtype is not included in the list
and, because Link itself succeeded, the failure is silent in the UI
(same UX gap as #1792).
Add the three subtypes to Loan::SUBTYPES, and add a regression test
asserting every subtype emitted by TYPE_MAPPING is valid for its
accountable so the mapper and models can't drift apart again.
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* fix(sharing): scope import account selects to accessible_by (#1803)
Three CSV / QIF account selects in import/uploads/show.html.erb and one
PDF-import account select in imports/_pdf_import.html.erb pulled their
options from `@import.family.accounts`. That listed every account in
the family — including the family admin's unshared personal accounts —
in the dropdown shown to any member running an import. Swap each call
site to `Current.user.accessible_accounts` (owned + explicitly shared
accounts only), matching the existing scoping used by the dashboard
sidebar, transactions controller, transfers controller, etc.
Adds a regression test that signs in as family_member and asserts the
unshared-account names from the dylan_family fixtures never appear in
the rendered upload page.
* test(import): scope leak assertions to account select (#2194 CodeRabbit)
CodeRabbit nitpick: assert_match on response.body could pass/fail on
text outside the account dropdown (sidebar, breadcrumb, error message,
etc.) and gave false confidence in the refute_match exclusions. Switch
to assert_select 'select[name="import[account_id]"] option', text: …
so the assertions only see the option nodes the leak test actually
cares about.
* test(import): cover PDF account-select scoping; pluck PDF partial (#2194 review)
jjmata: the _pdf_import.html.erb scoping change was not covered by the
existing test (which only hit /import/uploads). Add a regression test
hitting GET /imports/:id with a PdfImport fixture, asserting the
account dropdown options match accessible accounts only.
Also swap the PDF partial's accounts.map { |a| [a.name, a.id] } for
.pluck(:name, :id) to match the .pluck pattern the other three CSV/QIF
selects already use.
* test(import): stub pdf_uploaded? on PDF leak test (#2194 ci)
The new regression test hit ImportsController#show which redirects to
the upload page when @import.pdf_uploaded? is false. The pdf_with_rows
fixture has neither a pdf_file attached nor a statement, so the
redirect fired before the partial under test ever rendered, failing
with 302 in CI. Stub PdfImport#pdf_uploaded? to true so the test
exercises the account-select scoping path it was written to cover.
* fix(import): scope PDF form to :import so field names match (#2194 ci)
The PDF-import account-select form was `form_with model: import` with
no explicit scope. Because the model is a PdfImport, Rails derived the
param namespace from the class name, so the rendered field was named
`pdf_import[account_id]` — not `import[account_id]`. The
ImportsController#update action accepts either via
`params.dig(:pdf_import, :account_id) || params.dig(:import,
:account_id)` so live submissions still worked, but the regression
test added in 0685bbdf asserted on `select[name="import[account_id]"]`
and matched zero options.
Add `scope: :import` to align the rendered name with both the test
selector and the convention used by the CSV/QIF forms on the upload
page (which all use `scope: :import`).
* feat(mcp): add OAuth well-known discovery endpoints (RFC 8414 + RFC 9728)
Serves /.well-known/oauth-protected-resource (RFC 9728) and
/.well-known/oauth-authorization-server (RFC 8414) so MCP clients
can auto-discover the authorization server. Both endpoints are
unauthenticated and respect APP_URL for reverse-proxy deployments.
* feat(mcp): add dynamic client registration endpoint (RFC 7591)
POST /register creates a public Doorkeeper::Application on demand so
MCP clients (e.g. Claude.ai) can self-register without manual setup.
Validates redirect_uris (including blank entries), falls back to
"MCP Client" name, returns no client_secret (public client, PKCE only).
Rate-limited to 10 registrations/min/IP via Rack::Attack.
* feat(mcp): authenticate via Doorkeeper OAuth2, keep MCP_API_TOKEN as fallback
MCP endpoint now accepts OAuth2 Bearer tokens issued by Doorkeeper.
Falls back to the existing MCP_API_TOKEN env-var flow so self-hosted
deployments are not broken. Requires MCP_OAUTH_ENABLED or MCP_API_TOKEN
to be set — the endpoint returns 503 otherwise.
- OauthBase concern provides APP_URL-aware configured_base_url (trailing
slash stripped to prevent double-slash URLs)
- Bearer scheme parsed case-insensitively (RFC 7235)
- Only read_write scope accepted — read scope would allow mutating tools
(CreateGoal, ImportBankStatement), so read-only tokens are rejected
- Deactivated users rejected even with a valid Doorkeeper token
- WWW-Authenticate header on 401 points to RFC 9728 resource metadata
- SHA-256 digest used for constant-time env-var comparison
- Rack::Attack throttle added for POST /register
- Routes wired: /.well-known/*, /register, use_doorkeeper
* fix(mcp): disable Turbo on OAuth consent form for external redirect URIs
Turbo was intercepting the authorization form POST and XHR-fetching
the redirect_uri (e.g. https://claude.ai/api/mcp/auth_callback),
which CORS blocks. Extend the existing turbo_disabled guard to cover
any redirect_uri that doesn't originate from the app itself.
* feat(mcp): add Settings::McpController with connected clients view
- Settings > MCP page (under Advanced) shows the MCP server URL with
copy button and step-by-step instructions for connecting Claude.ai
- Lists active non-mobile OAuth tokens with app name and revoke action;
mobile device tokens are excluded to prevent accidental disconnection
- Removes the MCP_OAUTH_ENABLED env-var gate — OAuth auth is always
available since Doorkeeper handles consent; MCP_API_TOKEN remains
as a self-hosted fallback
* fix(mcp): remove client_credentials from grant_types_supported metadata
Only authorization_code is supported by the registration endpoint.
Advertising client_credentials was misleading — a client that reads
the metadata and attempts that flow would get an application with the
wrong grant type.
* fix(imports): import Actual rows with blank payee
Actual Budget exports reconciliation and starting-balance rows with a
blank Payee. ActualImport mapped the row name straight from the Payee
column with no fallback (unlike Import and MintImport), so a blank Payee
produced a blank Entry name. Entry requires a name, and import! wraps all
rows in a single transaction, so one blank-payee row failed validation
and rolled back the entire import -- surfacing only a generic "Import
failed" while the worker logged "done".
Fall back to the Notes column (which carries text like "Reconciliation
balance adjustment") and then to the default row name, matching the
blank-name handling already used by the base importer and MintImport.
Add a blank-payee row to the Actual fixture and regression tests covering
the Notes fallback, the default fallback, and an end-to-end import that
no longer fails on blank-payee rows.
* ci(security): skip calendar-based brakeman Rails EOL check
The scan_ruby job fails because brakeman's CheckEOLRails warns that Rails
7.2.3.1 reaches end of life on 2026-08-09. That check fires purely on the
calendar -- it warns 60 days before the EOL date and escalates in
confidence as the date nears (brakeman/checks/eol_check.rb) -- so it turns
`bin/brakeman` (exit code 3) red on every branch and on main regardless of
the code being scanned.
Add config/brakeman.yml (auto-loaded by `bin/brakeman`) skipping only
CheckEOLRails. CheckEOLRuby is left enabled because the current Ruby is not
near end of life, so that signal is preserved. A TODO records that the skip
should be removed when Sure upgrades off Rails 7.2.
Auto-categorization, merchant detection/enhancement, and PDF/bank-statement
extraction hard-coded Provider::Registry.get_provider(:openai), so selecting
Anthropic (or running an Anthropic-only self-hosted install) left those
operations using/missing OpenAI rather than the chosen provider.
Add Provider::Registry.preferred_llm_provider, which resolves the LLM provider
honoring Setting.llm_provider with a configured-provider fallback (mirroring how
chat picks its provider), and route all six TODO(#2113) call sites through it:
- Family::AutoCategorizer#llm_provider
- Family::AutoMerchantDetector#llm_provider
- ProviderMerchant::Enhancer#llm_provider
- PdfImport (process_pdf + extract_bank_statement)
- Assistant::Function::ImportBankStatement
Provider::Anthropic already implements auto_categorize / auto_detect_merchants /
enhance_provider_merchants (#1984) and process_pdf / extract_bank_statement
(#1985), so no provider changes are needed — only the wiring.
Closes#2113.
* feat(ai): add Anthropic provider with chat parity (1/5)
Introduces Provider::Anthropic alongside Provider::Openai, implementing
the LlmConcept chat_response contract over the official anthropic Ruby
SDK. Batch ops, PDF, and RAG land in follow-up PRs.
- Provider::Anthropic uses Messages API for sync and streaming responses
- ChatConfig builds requests with ephemeral prompt-cache markers on the
system prompt and the last tool definition
- MessageFormatter reconstructs multi-turn history (text + tool_use +
tool_result blocks) from raw Message records, including the paired
user-role tool_result turn Anthropic requires after every tool_use
- ChatParser maps Anthropic Message into the shared ChatResponse Data
- Registry, Setting, User, Chat default model wired for ANTHROPIC_*
envs and Setting.anthropic_*; LLM_PROVIDER selects between providers
- Responder forwards raw conversation_history (Array<Message>) so
providers without hosted conversation state can rebuild context
- OpenAI provider accepts and ignores the new kwarg (no behavior change)
Tests cover provider init, model gating, MessageFormatter for all turn
shapes, ChatConfig request building (max_tokens, system cache, tool
conversion), ChatParser for text / tool_use / mixed blocks, Registry
discovery, and mocked chat_response success / error / function_request
paths. Live VCR cassettes recorded in a follow-up with a real key.
Stacked PRs: 2/5 batch ops + cost ledger, 3/5 PDF, 4/5 pgvector RAG,
5/5 settings UI + disclosure.
* fix(ai): address PR review on Anthropic provider foundation
Surface fixes raised by Codex + CodeRabbit on PR 1/5:
- Provider::Anthropic#chat_response now accepts (and ignores) a
`messages:` kwarg. Assistant::Responder passes both `messages:`
(OpenAI-shape) and `conversation_history:` (raw Message records) for
cross-provider parity, so the previous signature raised
ArgumentError on the first chat turn through the Anthropic provider.
- Provider::Anthropic#supports_model? bypasses the `claude` prefix
gate when a custom base_url is configured, mirroring the OpenAI
provider. Bedrock-shaped IDs like
`anthropic.claude-sonnet-4-5-20250929-v1:0` and
`claude-opus-4@20250514` are otherwise rejected by
Assistant::Provided#get_model_provider and the chat dies.
- Setting.anthropic_access_token is now in
EncryptedSettingFields::ENCRYPTED_FIELDS so the Anthropic API key
is encrypted at rest like every other provider secret. Previously
plaintext while siblings (openai_access_token, twelve_data_api_key,
external_assistant_token) were ciphertext.
- Chat.default_model falls back to whichever provider is actually
configured. Previously, with LLM_PROVIDER=anthropic but no
Anthropic credentials, the default model resolved to a Claude ID
that no registered provider supported, so chats failed even when
OpenAI was fully configured. Adds Provider::{Anthropic,Openai}#configured?
class methods for the readable callsite.
- Provider::Anthropic.effective_model uses
`ENV["ANTHROPIC_MODEL"].presence || Setting.anthropic_model` so the
Setting lookup is only performed when the env var is absent — the
previous `ENV.fetch(KEY, default)` evaluated the default arg
eagerly on every call.
- Provider::Anthropic::ChatConfig#anthropic_input_schema strips both
`:strict` and `"strict"` keys so JSON-decoded schemas with string
keys cannot leak the OpenAI-only flag through to Anthropic.
Test coverage added: supports_model? bypass on custom endpoints,
chat_response messages: kwarg compatibility, default_model fallback
in the three credential combinations, configured? against ENV +
Setting, strict-flag stripping for both key types, and a
`Setting.expects(:anthropic_model).never` assertion proving the
ENV-precedence test now exercises the lazy path.
All 4365 tests pass (1 pre-existing libvips env error unrelated).
* test(chat): make default_model tests resilient to ENV model overrides
CodeRabbit flagged on PR review: the new default_model tests asserted
against Provider::*::DEFAULT_MODEL, but Chat.default_model actually
returns Provider::*.effective_model.presence (which reads
OPENAI_MODEL / ANTHROPIC_MODEL from the environment). With either env
var set, the tests would fail intermittently even though routing was
correct.
- New default_model tests now assert against the provider's
effective_model directly, so they verify the routing decision
(which provider's value wins) without coupling to the constant.
- Pre-existing "creates with default model" assertions had the same
brittleness; switch them to compare against Chat.default_model so
the chosen model is whatever the env / Setting cascade resolves to.
Verified by running `ANTHROPIC_MODEL=claude-haiku-4-5 OPENAI_MODEL=gpt-4o
bin/rails test test/models/chat_test.rb` — 16 runs, 0 failures
(previously 2 pre-existing failures + 0 from the new tests).
* fix(ai): address local review on Anthropic foundation
- Provider::Anthropic#supports_pdf_processing? bypasses prefix gate for
custom endpoints, mirroring supports_model?
- Provider::Anthropic#initialize raises Error when custom_endpoint? AND
model.blank?, parity with Provider::Openai
- stream_chat_response captures partial usage on mid-stream errors and
records it via the new on_partial callback so chat_response can skip
the duplicate error row in the outer rescue
- safe_accumulated_message swallows the secondary failure when the SDK
cannot reconstruct a snapshot
- langfuse_client memoizes properly (||= instead of =) so repeated calls
don't churn Langfuse instances
- MessageFormatter sorts tool_calls by created_at then id so the
message array is deterministic across replays; skips tool_calls
missing both provider_call_id and provider_id rather than sending
`id: nil` and getting rejected by Anthropic
- Setting.anthropic_access_token default falls back through
ENV["ANTHROPIC_API_KEY"].presence (was missing .presence, so an
empty-string env value bled through)
- User#openai_configured? / #anthropic_configured? delegate to the
Provider::* class methods — single source of truth
- Assistant::Responder renames the OpenAI-shape history builder
conversation_history → openai_messages_payload so the kwarg name
matches the local method name (messages: openai_messages_payload,
conversation_history: chat_message_records)
- Assistant::Builtin stale-history comment updated to reference both
builders
Adds a streaming chat_response test using ad-hoc subclasses of the
SDK event types so the case/when dispatch matches via is_a? without
stubbing class-level === behavior.
* test(ai): add Anthropic tool_use round-trip + multi-tool turn coverage
Addresses @jjmata's "worth confirming" note on PR #1983: tool-use turns
from prior assistant messages must round-trip correctly when retrieved
from the database.
- New `ChatParser → ToolCall::Function → MessageFormatter` test walks
the full path: Anthropic response with a tool_use block →
ChatFunctionRequest → ToolCall::Function.from_function_request →
persisted on the AssistantMessage → MessageFormatter rebuild on the
next turn. Asserts the original `tool_use.id` is preserved end-to-end
as both `tool_use.id` and the paired `tool_result.tool_use_id`, and
that the original `input` hash and serialized result content survive.
- New multi-tool assistant turn test confirms two tool_use blocks on a
single assistant message render as two tool_use blocks followed by
two paired tool_result blocks in a single user-role follow-up,
matching Anthropic's required alternation.
Both tests exercise the existing PR1 code without behavior changes.
* test(ai): require "ostruct" explicitly in Anthropic provider tests
OpenStruct is moving out of Ruby's default load path (warning in 3.4+,
removed in 3.5+). Tests work today because ActiveSupport transitively
loads it, but that's incidental. Match the existing convention in
test/controllers/settings/hostings_controller_test.rb which explicitly
requires ostruct for the same reason.
* fix(ai): sanitize Langfuse warn logs, normalize tool_use.input, dedup history fetch
Addresses three open CodeRabbit findings on PR #1983.
- Provider::Anthropic Langfuse rescue branches no longer include
`e.full_message` in `Rails.logger.warn`. `full_message` bundles the
backtrace + cause chain and on some SDK error types includes the
serialized request/response payload (prompt, model output). Logs
now report `#{e.class}: #{e.message}` only. Three sites:
create_langfuse_trace, log_langfuse_generation, upsert_langfuse_trace.
Note: Provider::Openai has the same pattern (copy-pasted source) —
harmonization deferred to a follow-up cleanup PR; this commit fixes
only the Anthropic provider to keep PR scope tight.
- MessageFormatter#parse_arguments now coerces any non-Hash parsed
result to `{}`. Anthropic's Messages API requires `tool_use.input`
to be a JSON object (map); a stored ToolCall::Function record whose
arguments parse to a scalar, bool, or array (corrupt row, legacy
data, cross-provider bleed) would otherwise produce a payload the
API rejects. Normal flow stores Hash arguments end-to-end so the
fix is defensive — adds 2 tests covering scalar/array JSON strings
and non-String non-Hash inputs.
- Assistant::Responder dedups the chat-history fetch. The previous
layout fired two near-identical `chat.messages.where(...).includes(
:tool_calls).ordered` queries per LLM turn (one for the OpenAI-shape
payload, one for the raw-records kwarg). A new memoized
`complete_chat_messages` fetches once; `chat_message_records` filters
out the current message via `Array#reject`, `openai_messages_payload`
iterates the cached array unchanged. One SQL query per turn instead
of two. Memoization scope = single Responder instance (per LLM call),
so cache invalidation is not a concern.
All 4370 tests pass (1 pre-existing libvips env error unrelated).
Rubocop + brakeman clean.
* fix(ci): replace sk-ant- prefixed test placeholders
Pipelock secret scanner pattern-matches `sk-ant-*` as a real Anthropic
API key and fails the PR security-scan check. Test stubs and
ClimateControl env values used `sk-ant-test`, `sk-ant-from-setting`,
`sk-ant-x`, `sk-ant-y` as obvious placeholders, but the scanner does
not care about value entropy.
Switched to `fake-anthropic-key-*` / `fake-token-*` strings so the
scanner stops flagging them. No production code touched, no behavior
change — Provider::Anthropic still accepts any non-blank token.
* feat(ai): add Anthropic batch ops + LLM cost ledger (2/5)
Implements auto_categorize, auto_detect_merchants, and
enhance_provider_merchants on Provider::Anthropic via forced tool calls,
plus the cost-ledger plumbing they need.
- Provider::Anthropic::AutoCategorizer, AutoMerchantDetector,
ProviderMerchantEnhancer each define a single output tool whose
input_schema mirrors the desired output, then force the model to call
it via tool_choice: { type: "tool", name: ..., disable_parallel_tool_use: true }.
Anthropic guarantees the tool_use.input matches the schema, so there
is no JSON parsing fragility, no <think> tag stripping, and no
json_object/json_schema fallback ladders.
- Concerns::UsageRecorder mirrors the OpenAI sibling but persists
cache_creation_input_tokens / cache_read_input_tokens to dedicated
columns instead of metadata.
- Migration adds cache_creation_tokens, cache_read_tokens (nullable
integers) to llm_usages. OpenAI rows leave them null.
- LlmUsage::PRICING gains Claude 4.x rows (opus-4-7 $15/$75, sonnet-4-6
$3/$15, haiku-4-5 $1/$5 per MTok). infer_provider returns "anthropic"
for claude-* via the existing exact/prefix lookup.
- Provider::Anthropic#chat_response now persists cache columns directly
rather than stashing them in metadata.
- 25-transaction batch cap mirrors the OpenAI provider so the cost
ledger sees the same shape regardless of which provider ran a batch.
Tests cover the forced-tool-call path, null/None normalization,
case-insensitive merchant matching, the missing-tool_use error path,
and Anthropic-specific pricing + provider inference on LlmUsage.
Stacked on #1983 (PR 1/5). 3/5 PDF + vision next.
* fix(ai): attribute Bedrock model IDs to anthropic + clean nil enum
- LlmUsage.infer_provider now returns "anthropic" for Bedrock /
Vertex shaped IDs (anthropic.* and anthropic/*), so cost-ledger
filtering by provider stays correct even when no per-MTok rate is
stored. Previously these IDs fell through to the "openai" default.
- AutoCategorizer drops the redundant nil sentinel from the
category_name enum — the union type [string, null] already permits
null, and some JSON Schema validators reject nil literals inside
enum arrays.
* test(ai): require "ostruct" in Anthropic batch op tests
Same rationale as the PR1 ostruct fix — explicit require so the tests
don't depend on ActiveSupport's transitive load when Ruby 3.5+ removes
OpenStruct from the default load path.
* feat(ai): Anthropic native PDF processing (3/5)
Implements process_pdf and extract_bank_statement on Provider::Anthropic
using the native `document` content block — no rasterization, no text
pre-extraction.
- Provider::Anthropic::PdfProcessor classifies the document, summarizes
it, and extracts statement metadata via a forced report_document_analysis
tool whose input_schema mirrors the existing Provider::Openai output
(document_type from Import::DOCUMENT_TYPES, summary, extracted_data).
- Provider::Anthropic::BankStatementExtractor returns the same
{ transactions, period, account_holder, account_number, bank_name,
opening_balance, closing_balance } shape via report_bank_statement so
downstream pdf_import code is provider-agnostic.
- Both attach the PDF as
{ type: "document", source: { type: "base64", media_type: "application/pdf", data: <b64> } }
— Claude 3.5+ / 4.x accept this natively (up to 32MB / 100 pages).
No pdf-reader, no pdftoppm, no chunking for typical statements.
- supports_pdf_processing? (introduced in PR 1) already returns true for
claude-* models, gating process_pdf with a clear error otherwise.
- Cost ledger rows are persisted via the shared UsageRecorder concern,
including cache_creation/cache_read tokens.
Tests verify the document block shape, tool_choice forcing, normalized
document_type for unknown classifications, transaction normalization
(date / amount / reference → notes), and the missing-tool_use error
path. Blank pdf_content raises before any client call.
Stacked on #1984 (PR 2/5). 4/5 pgvector RAG next.
* fix(ai): guard PDF size + surface bank-statement truncation
- PdfProcessor and BankStatementExtractor raise upfront when
pdf_content.bytesize exceeds MAX_PDF_BYTES (32 MB, matching
Anthropic's hard limit). Previously a 100 MB PDF would be
base64-encoded (~133 MB) and packed into the JSON body before
the API rejected it — peak heap ~270 MB per Sidekiq worker.
- BankStatementExtractor inspects response.stop_reason; when the
model hit max_tokens it logs a warning and flags result[:truncated]
so downstream callers know the transaction list may be incomplete.
- ISO date pattern added to statement_period_start/end schema in
PdfProcessor so the model can't return "March 2026" — Anthropic
enforces the regex via the tool's input_schema.
Tests cover the size guard (raises before any client.messages call),
truncated-result flagging, and the warning log path.
* test(ai): require "ostruct" in Anthropic PDF tests
Match the explicit ostruct require added in PR1/PR2 — same Ruby 3.5+
load-path reason.
* feat(ai): default Anthropic installs to pgvector RAG (4/5)
The provider-agnostic vector store stack (VectorStore::Pgvector + the
Embeddable concern) already shipped to main. This PR closes the
Anthropic loop:
- VectorStore::Registry.adapter_name now returns :pgvector when
Setting.llm_provider == "anthropic" and no explicit
VECTOR_STORE_PROVIDER override is set. Anthropic has no hosted vector
store, so falling back to the local pgvector adapter is the only
correct default. Explicit VECTOR_STORE_PROVIDER still wins.
- SearchFamilyFiles surfaces a longer message when no adapter is wired
up — calling out pgvector + EMBEDDING_URI_BASE as the supported
Anthropic-only path so the user is not stuck with an "OpenAI required"
hint that is no longer accurate.
The Embeddable concern already pulls embeddings from
EMBEDDING_URI_BASE / EMBEDDING_ACCESS_TOKEN (with OpenAI as fallback),
so Anthropic installs point this at Voyage AI, a local Ollama instance,
or OpenAI embeddings — independent of the chat provider.
Tests cover the new default routing, the existing OpenAI default
staying intact, and explicit VECTOR_STORE_PROVIDER overriding the
Anthropic default.
Stacked on #1985 (PR 3/5). 5/5 settings UI + retention disclosure next.
* feat(ai): self-host settings UI for Anthropic provider (5/5)
Adds the Anthropic panel and the install-wide LLM provider selector to
the self-hosting settings page, plus a shared data-retention
disclosure that covers both OpenAI and Anthropic.
- New _llm_provider_selector partial: select for Setting.llm_provider
(openai | anthropic), respects the LLM_PROVIDER env var (disables the
control + shows the "configured through environment variables" hint
when set, mirroring the existing OpenAI panel behaviour), and renders
a compact data-handling block with one-line retention statements for
each provider.
- New _anthropic_settings partial mirrors _openai_settings exactly:
password-field for the API key with **** redaction, optional
base_url (for AWS Bedrock / GCP Vertex), optional default model. All
three fields disable when their ENV var is set.
- show.html.erb renders provider selector + OpenAI panel + Anthropic
panel under the same "General" section so users can configure either
(or both) without switching pages.
- Settings::HostingsController#update now permits and persists
anthropic_access_token (ignoring the **** placeholder, same pattern
as OpenAI), anthropic_base_url, anthropic_model, and llm_provider
(validated against %w[openai anthropic]). On Setting::ValidationError
the rescue branch preserves anthropic_base_url / anthropic_model
input so the form re-renders with the user's typed values intact —
parity with the issue #1824 fix for OpenAI.
- Locale keys added under settings.hostings.{llm_provider_selector,
anthropic_settings}.
Tests cover token update + placeholder redaction, base_url + model
update, llm_provider switch to anthropic, and rejection of unknown
provider values. The existing GET render test still passes, exercising
all three new partials.
Closes the 5/5 Anthropic series stacked on #1986.
* fix(ai): valid Tailwind token + base_url URL validation
- Data-handling block in _llm_provider_selector swaps the invalid
bg-surface-secondary token for bg-container-inset, matching the
inset-card pattern used elsewhere in sure-design-system/components.css.
bg-surface-secondary is not defined anywhere in the design system —
Tailwind treated it as a no-op, so the block rendered with no
background contrast.
- Settings::HostingsController validates anthropic_base_url as a
URI::HTTP (catches https too) and raises Setting::ValidationError
with a localized message when the input is not parseable.
Previously any string was persisted, surfacing as an opaque
connection error at request time instead of an immediate UX failure.
- Blank base_url now clears the setting (was already the case but
exercised explicitly in tests now).
* fix(ci): replace sk-ant- prefixed token in hostings controller test
Same pipelock secret-scan trigger as PR1 fix on registry/anthropic
tests. The sk-ant-* prefix is matched verbatim by the scanner
regardless of value entropy.
* fix(ai): provision pgvector table when it is the default store
#1986 makes pgvector the default vector store for Anthropic installs, but
CreateVectorStoreChunks only ran when VECTOR_STORE_PROVIDER=pgvector was set
explicitly — so a fresh Anthropic-only install migrated without the
vector_store_chunks table and failed on uploads/searches.
Add VectorStore::Registry.pgvector_effective? as the single source of truth
for "is pgvector active?" (explicit env OR the Anthropic default), and a new
idempotent migration that enables the extension + creates the table whenever
pgvector is effective and the table is missing — covering fresh and
already-migrated installs without drift. Addresses Codex P1.
* fix(ai): provision pgvector table for Anthropic-default installs
Migration gated on raw VECTOR_STORE_PROVIDER==pgvector, so an
Anthropic-default install (which selects pgvector implicitly via
Setting.llm_provider without setting VECTOR_STORE_PROVIDER) skipped
table creation and failed later on a missing vector_store_chunks
relation. Route through VectorStore::Registry.pgvector_effective? —
the single source of truth already shared by the adapter selection.
Addresses Codex P1 review finding.
* fix(ai): provision pgvector chunks table on schema-load installs
The ensure-migration only helps db:migrate upgraders. Fresh installs go
through bin/docker-entrypoint's db:prepare, which loads schema.rb (the
conditional table can't be dumped there — it needs the vector extension)
and marks every migration applied without running it. An Anthropic-only
fresh install therefore selected the pgvector adapter but had no table,
failing with raw PG errors on first upload or search.
Two layers close it:
- VectorStore::Pgvector#ensure_schema! provisions the table idempotently
on first use (mirrors CreateVectorStoreChunks; memoized; failures wrap
in VectorStore::Error, which with_response turns into a clean failed
response).
- VectorStore::Registry#build_pgvector now gates on
VectorStore::Pgvector.available? (table exists, or extension present),
so installs whose Postgres lacks pgvector entirely degrade to the
assistant's provider_not_configured message instead of raising
mid-chat.
Also resolves the schema.rb version conflict against main (keep the
branch's 2026_06_01_120000, on top of main's current tables).
* fix(ai): address review nitpicks on pgvector provisioning
- Registry: update the adapter doc comment to mention the
Anthropic-to-pgvector default alongside the openai fallback.
- ensure_schema!: guard the DDL with if_not_exists instead of a Mutex.
Adapter instances are built per call and never shared across threads,
so the realistic race is two processes (web + Sidekiq) provisioning
concurrently; IF NOT EXISTS makes the loser a no-op where a Mutex
would only serialize threads inside one process.
* fix(ai): address review on Anthropic settings UI
- Require an Anthropic model when a custom base URL is saved, mirroring the
OpenAI branch. Auto-submit-on-blur could persist a base URL with no model,
making Provider::Anthropic raise "Model is required..." on every LLM call.
- Narrow the LLM provider selector copy: only chat honors Setting.llm_provider;
categorization, merchant detection and PDF processing still always use OpenAI.
Stop advertising provider switching for those flows until they are wired.
- Reset global Setting.* in test teardown to prevent state leakage, and add a
test covering the new base-URL-requires-model validation.
* feat(ds): conditional LLM provider settings + merged copy
The self-hosting AI section showed both providers' credential blocks at once
and duplicated near-identical copy. Tidy it:
- Replace the provider <select> with a DS::SegmentedControl driving a new
provider-settings Stimulus controller: only the active provider's panel is
shown; switching reveals the other instantly and persists Setting.llm_provider.
- Merge the two byte-identical data-retention lines into one provider-neutral
Data handling note.
- Scope the token-budget copy to OpenAI-compatible calls (read only by
Provider::Openai) and add an inline 'add a key to activate' hint when the
active provider is unconfigured.
UI-only; no provider behavior change.
* feat(ds): responsive LLM provider picker (tabs >=sm, select on mobile)
The segmented tabs overflow a phone viewport once there are 3+ providers
(measured: 4 labels want ~409px in a 319px column at 390px wide). Below sm,
fall back to a native <select> -- which doubles as the submitted field -- while
keeping the segmented tabs at sm and up.
Both controls bind to the same provider-settings Stimulus controller (the
select reads its value, the tabs read data-provider), so adding a 3rd/4th
provider scales on mobile with no layout math.
* fix(hostings): sanitize llm provider selector
* test(hostings): avoid brittle provider hint assertion
---------
Co-authored-by: sure-admin <sure-admin@splashblot.com>
