sax3l/accounted
3
0
Fork 0
Code Issues 1 Pull Requests 2 Actions Packages Projects Releases Wiki Activity
1,508 Commits 6 Branches 0 Tags
c0ecf2fa3bebd46bdfd0169efd73b89653d1dfed
Commit Graph

3 Commits

Author SHA1 Message Date
Mattsson 84c8e1ce59 fix(inbox): never extract the receiving company as its own supplier (#2080) 
* fix(inbox): never extract the receiving company as its own supplier

The model sometimes reads the Kund/Kunduppgifter block of bank agreements
and similar documents as the issuer, so the inbox offered to create the
user's own company as a leverantor. Two layers: a supplier-direction rule
in the extraction prompt (issuer, never the customer/recipient block), and
a deterministic post-extraction guard that nulls the supplier block when
its org number, derived VAT number, or exact name equals the receiving
company's own. All AI extraction paths (upload, deferred, retry,
attach-document, document-extraction, MCP) pass the company identity.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Hqm9QgdyNAFaWiz6Ww7pgb

* fix(inbox): keep the own-company guard alive for photographed documents

Skeptic findings on PR #2080: normalizeImageForExtraction rebuilt the input
without ownCompany, so the guard never fired for HEIC/oversized phone photos
(the motivating case). Spread the original input instead. Also let a
provably different extracted org number outvote a name coincidence, and
accept 12-digit personnummer-form own org numbers (19/20 century prefix,
enskild firma) alongside the 16-prefixed organisation form.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Hqm9QgdyNAFaWiz6Ww7pgb

* docs(compliance): add RoPA entry for the own-company-identity lookup

Compliance swarm (GDPR Art. 30): fetchOwnCompanyIdentity reads
companies.name and org_number on every extraction; record the processing
activity (transient in-memory comparison, never sent to the model).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Hqm9QgdyNAFaWiz6Ww7pgb

* fix(inbox): log own-company identity lookup failures instead of failing silent

Compliance swarm (SOC 2 CC7.2): the bare catch in fetchOwnCompanyIdentity
made a persistently broken lookup (RLS misconfig, DB outage) disable the
guard invisibly. Also pin the VAT-number match across prefixed and bare
digit forms in the test matrix.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Hqm9QgdyNAFaWiz6Ww7pgb

* fix(inbox): surface maybeSingle query errors in the own-company lookup

CodeRabbit: maybeSingle() reports RLS/query failures in error without
throwing, so the failure log added for SOC 2 CC7.2 never fired for exactly
those cases. Throw the reported error into the existing catch (still fail
open to nulls).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Hqm9QgdyNAFaWiz6Ww7pgb

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-31 15:13:14 +02:00
Jakob Wennberg 524d9978f1 fix(migration): resumable underlag import without inline extraction + same-origin MCP storage URLs (#1783) 
* fix(migration): resumable underlag import without inline extraction, same-origin MCP storage URLs

The Fortnox underlag import ran every file's AI extraction inline inside
one request and hit the hosted 300 s function limit after ~17 of 113 files
(twice on 2026-08-21); the UI showed the generic "underlagen kunde inte
importeras" although the files it did reach were linked. The import now
works in time-budgeted slices with a stable cursor (the UI loops until the
server reports the end and shows "x av y") and opts out of extraction
(extractionOwner 'none', stamped skipped:opted_out): every file is linked
to its posted verifikat on arrival, so the booking is already known.

MCP signed Storage URLs (upload_url, signed_url, download_url) are served
through a same-origin proxy, /api/storage/[...path], because Claude
Desktop's sandbox only reaches the MCP host and blocked the PUT to
<project>.supabase.co. The signed token stays the only credential; the
proxy forwards only signed documents-bucket paths to our own Storage host
and is a no-op rewrite when NEXT_PUBLIC_APP_URL is unset.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013YoZ8iboyTj221axW6Gdtm

* fix(mcp): keep the storage-proxy note out of the size-capped tool descriptions

The per-tool 280-char cap and the tools/list payload ceiling both tripped on
the two sentences added to gnubok_create_document_upload and
gnubok_get_document_content; the why now lives in a code comment.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013YoZ8iboyTj221axW6Gdtm

* fix(review): id cursor, stall = error, capped upload body, encoded dot segments

Review follow-ups on #1783:
- the import cursor is the last handled provider attachment id, not an
  index, so a file Fortnox adds or removes mid-sweep shifts nothing
- a partial answer whose cursor does not advance (or the round guard) is
  reported as ARCIM_DOCUMENT_IMPORT_STALLED instead of "complete"; the
  slices already landed stay reported and the retry button resumes
- the storage proxy reads the PUT body as a capped stream instead of
  buffering an unbounded payload before measuring it
- object paths are rejected when any segment decodes to "." or ".." (or
  holds a separator), and the URL fetch() would actually request is
  re-checked against the allowlist after normalisation
- download_url description no longer claims a direct Storage URL

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013YoZ8iboyTj221axW6Gdtm

---------

Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-21 15:28:37 +02:00
Jakob Wennberg c7a75d069d feat(ai): job-shaped AI service with OpenAI-compatible backend, extraction-first; stop extracting every inbox document twice (#1740) 
* feat(ai): job-shaped AI service with OpenAI-compatible backend, extraction-first; stop extracting every inbox document twice

Sovereign plan WS1 PR1 (#1406 Tier 2, extraction-first, aligned with the
AI surface audit).

lib/ai grows a job-shaped service (generateText / generateStructured /
extractFromDocument; no streaming members yet, see plan rule R3):
- services/anthropic-family delegates to the existing createAiClient()
  and sends the exact request literals the inbox extractor sent before
  (request-shape tests deep-equal them), so hosted Bedrock stays
  byte-identical.
- services/openai-compatible talks to any chat-completions endpoint
  (BYO Swedish provider) via Vercel AI SDK 6.x, exact-pinned and
  guarded: images as parts, PDFs rasterized with poppler (AI_PDF_MODE)
  or sent natively, AI_VISION / AI_STRICT_JSON declared, honest skips
  (ai_no_vision, pdf_rasterizer_missing) instead of fake failures.
- config.ts: AI_PROVIDER/AI_BASE_URL/AI_API_KEY/AI_MODEL and per-tier
  AI_*_MODEL with the legacy BEDROCK_* names kept as the same overrides;
  getAiStatus() is the single source of truth for "is AI wired up".
- provider.ts: openai-compatible in the auto-detect chain (after Bedrock
  and the direct API); createAiClient() refuses it loudly.

Document extraction moves onto the service and gets the audit's fixes:
- Inbox documents were extracted TWICE (pipeline A ran inside
  uploadDocument() before the inbox row existed, so its dedupe branch
  never fired; 3 707 + 1 666 calls / 30 d). The inbox now declares
  extractionOwner on the upload, the extension yields, and the inbox
  mirrors its single outcome onto document_attachments from every
  writer (sync, deferred, attach, retry, MCP).
- Every "no extraction will ever happen" outcome is stamped
  (skipped:no_ai_entitlement / ai_unconfigured / system_generated /
  ...); the status route maps the quiet ones to 'disabled' on the first
  poll instead of a 30 s client timeout. Prod showed 309 of the 327
  never-extracted uploads were the paywall working silently.
- Self-generated documents (our own invoice PDFs, payout files) are no
  longer OCR'd.
- Agent invoke answers 503 ai_unconfigured when the deployment has no
  assistant backend, distinct from the paywall.

Guard: new direct-ai-client antipattern check (shrink-only allowlist of
the pre-abstraction SDK callers) plus exact pins for @anthropic-ai/sdk,
ai and @ai-sdk/openai-compatible.

Verified: 15 958 unit tests green, guards, lint ratchet, typecheck, and a
live smoke against hosted Bedrock through the new service (ping, streamed
tool turn, thinking+cache, PDF extraction).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* feat(ai): make AI_API_KEY optional for OpenAI-compatible endpoints (keyless local model servers)

A local model server (llama.cpp's server, Ollama /v1, LM Studio, vLLM)
usually has no auth. Before, the OpenAI-compatible backend required both
AI_BASE_URL and AI_API_KEY to count as configured, so running Accounted on a
local model meant setting a meaningless placeholder key.

- resolveAiProvider / hasAiCredentials: a base URL alone is now enough.
- services/openai-compatible: only send Authorization: Bearer when AI_API_KEY
  is set, so a keyless server is never handed an empty bearer; a hosted
  provider that needs a key still sets it.
- Docs (SELF-HOSTING Option 3: local-model example, key marked optional),
  DECISIONS.

Verified: with no AI_API_KEY, just AI_BASE_URL + AI_MODEL, getAiStatus()
reports configured=true / provider=openai-compatible (live). lib/ai suite
71 green; tsc, guards, lint clean. Bedrock/Anthropic logic unchanged.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
 2026-08-20 19:39:08 +02:00