feat(phase-b): ship analytics dashboard, alerts, scanner PWA, dedup, audit view
Phase B (Insights & Alerts) PR4-11 in one drop. Builds on the schema +
service skeletons committed in PRs 1-3.
PR4 Analytics dashboard — 4 chart types (funnel/timeline/breakdown/source),
date-range picker (today/7d/30d/90d), CSV+PNG export per card.
PR5 Alert rail UI + /alerts page — topbar bell w/ live count, dashboard
right-rail, three-tab page (active/dismissed/resolved), socket-driven
invalidation. Bell lazy-loads list on popover open to keep cold pages
fast in non-dashboard routes.
PR6 EOI queue tab on documents hub — filters to in-flight EOIs, count
surfaces in tab label.
PR7 Interests-by-berth tab on berth detail — replaces the stub.
PR8 Expense duplicate detection — BullMQ job runs scan on create, yellow
banner on detail w/ Merge / Not-a-duplicate, transactional merge
consolidates receipts and archives the source.
PR9 Receipt scanner PWA + multi-provider AI — port-scoped /scan route in
its own (scanner) group with no dashboard chrome, dynamic per-port
manifest, OpenAI + Claude provider abstraction, admin OCR settings
page (port-level + super-admin global default w/ opt-in fallback),
test-connection endpoint, manual-entry fallback when no key is
configured. Verify form always shown before save — no ghost rows.
PR10 Audit log read view — swap to tsvector full-text search on the
existing GIN index, cursor pagination, filters for entity/action/user
/date range, batched actor-email resolution.
PR11 Real-API tests — opt-in receipt-ocr.spec (admin save+test, optional
real-receipt parse via REALAPI_RECEIPT_FIXTURE) and alert-engine
socket-fanout spec gated behind RUN_ALERT_ENGINE_REALAPI. Both skip
cleanly without their gate envs so CI stays green.
Test totals: vitest 690 -> 713, smoke 130 -> 138, realapi +2 opt-in.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 17:21:55 +02:00
/ * *
* Receipt OCR provider adapters . Each adapter takes raw image bytes
* and returns a normalized ` ParsedReceipt ` shape ; callers don ' t care
* which provider produced it .
* /
import OpenAI from 'openai' ;
import { logger } from '@/lib/logger' ;
export interface ParsedReceiptLineItem {
description : string ;
amount : number ;
}
export interface ParsedReceipt {
establishment : string | null ;
/** ISO YYYY-MM-DD. */
date : string | null ;
amount : number | null ;
currency : string | null ;
lineItems : ParsedReceiptLineItem [ ] ;
/** 0..1; below 0.6 surfaces "verify mode" UI. */
confidence : number ;
}
feat(ai): per-port token budgets + usage ledger for AI features
Adds a token-denominated guardrail in front of every server-side AI call
so a misconfigured port can't run up an unbounded bill. Soft caps surface
a banner; hard caps refuse new requests until the period rolls over.
Usage flows into a feature-typed ledger so future AI surfaces (summary,
embeddings, reply-draft) can drop in without schema changes.
- New table ai_usage_ledger (port, user, feature, provider, model,
input/output/total tokens, request id) with two indexes for rollup
- New service ai-budget.service.ts: getAiBudget/setAiBudget,
checkBudget (pre-flight gate), recordAiUsage, currentPeriodTokens,
periodBreakdown — all token-based, period boundaries in UTC
- runOcr now returns provider usage so the route can record the actual
spend instead of estimating
- Scan-receipt route gates on checkBudget before invoking AI; returns
source: manual / reason: budget-exceeded when blocked, surfaces
softCapWarning on the success path
- Admin UI: new AiBudgetCard on the OCR settings page — shows current
spend, per-feature breakdown, soft/hard cap inputs, period selector
- Permission: admin.manage_settings on both routes
Tests: 766/766 vitest (was 756) — +10 budget tests covering enforce/
disabled/cap-exceed/estimate-exceed/soft-warn/period boundaries/
cross-port isolation/silent ledger failure.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 19:53:09 +02:00
export interface OcrUsage {
inputTokens : number ;
outputTokens : number ;
requestId : string | null ;
}
export interface OcrRunResult {
parsed : ParsedReceipt ;
usage : OcrUsage ;
}
feat(phase-b): ship analytics dashboard, alerts, scanner PWA, dedup, audit view
Phase B (Insights & Alerts) PR4-11 in one drop. Builds on the schema +
service skeletons committed in PRs 1-3.
PR4 Analytics dashboard — 4 chart types (funnel/timeline/breakdown/source),
date-range picker (today/7d/30d/90d), CSV+PNG export per card.
PR5 Alert rail UI + /alerts page — topbar bell w/ live count, dashboard
right-rail, three-tab page (active/dismissed/resolved), socket-driven
invalidation. Bell lazy-loads list on popover open to keep cold pages
fast in non-dashboard routes.
PR6 EOI queue tab on documents hub — filters to in-flight EOIs, count
surfaces in tab label.
PR7 Interests-by-berth tab on berth detail — replaces the stub.
PR8 Expense duplicate detection — BullMQ job runs scan on create, yellow
banner on detail w/ Merge / Not-a-duplicate, transactional merge
consolidates receipts and archives the source.
PR9 Receipt scanner PWA + multi-provider AI — port-scoped /scan route in
its own (scanner) group with no dashboard chrome, dynamic per-port
manifest, OpenAI + Claude provider abstraction, admin OCR settings
page (port-level + super-admin global default w/ opt-in fallback),
test-connection endpoint, manual-entry fallback when no key is
configured. Verify form always shown before save — no ghost rows.
PR10 Audit log read view — swap to tsvector full-text search on the
existing GIN index, cursor pagination, filters for entity/action/user
/date range, batched actor-email resolution.
PR11 Real-API tests — opt-in receipt-ocr.spec (admin save+test, optional
real-receipt parse via REALAPI_RECEIPT_FIXTURE) and alert-engine
socket-fanout spec gated behind RUN_ALERT_ENGINE_REALAPI. Both skip
cleanly without their gate envs so CI stays green.
Test totals: vitest 690 -> 713, smoke 130 -> 138, realapi +2 opt-in.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 17:21:55 +02:00
const EMPTY_RESULT : ParsedReceipt = {
establishment : null ,
date : null ,
amount : null ,
currency : null ,
lineItems : [ ] ,
confidence : 0 ,
} ;
const SYSTEM_PROMPT =
'You extract structured data from a marina-business receipt image. Return ONLY a JSON object with these keys: establishment (string), date (ISO YYYY-MM-DD), amount (number, total), currency (3-letter ISO code), lineItems (array of {description, amount}), confidence (number 0-1). If a field cannot be read, return null for that field. Set confidence near 0 if the image is unreadable, near 1 if every field was confidently extracted.' ;
interface RunArgs {
imageBuffer : Buffer ;
mimeType : string ;
apiKey : string ;
model : string ;
}
function safeParse ( content : string ) : ParsedReceipt {
const cleaned = content . replace ( /```json\n?|\n?```/g , '' ) . trim ( ) ;
try {
const obj = JSON . parse ( cleaned ) as Partial < ParsedReceipt > ;
return {
establishment : obj.establishment ? ? null ,
date : obj.date ? ? null ,
amount : typeof obj . amount === 'number' ? obj.amount : null ,
currency : obj.currency ? ? null ,
lineItems : Array.isArray ( obj . lineItems ) ? obj . lineItems : [ ] ,
confidence : typeof obj . confidence === 'number' ? obj.confidence : 0 ,
} ;
} catch ( err ) {
logger . warn ( { err , contentLen : cleaned.length } , 'OCR provider returned non-JSON' ) ;
return EMPTY_RESULT ;
}
}
feat(ai): per-port token budgets + usage ledger for AI features
Adds a token-denominated guardrail in front of every server-side AI call
so a misconfigured port can't run up an unbounded bill. Soft caps surface
a banner; hard caps refuse new requests until the period rolls over.
Usage flows into a feature-typed ledger so future AI surfaces (summary,
embeddings, reply-draft) can drop in without schema changes.
- New table ai_usage_ledger (port, user, feature, provider, model,
input/output/total tokens, request id) with two indexes for rollup
- New service ai-budget.service.ts: getAiBudget/setAiBudget,
checkBudget (pre-flight gate), recordAiUsage, currentPeriodTokens,
periodBreakdown — all token-based, period boundaries in UTC
- runOcr now returns provider usage so the route can record the actual
spend instead of estimating
- Scan-receipt route gates on checkBudget before invoking AI; returns
source: manual / reason: budget-exceeded when blocked, surfaces
softCapWarning on the success path
- Admin UI: new AiBudgetCard on the OCR settings page — shows current
spend, per-feature breakdown, soft/hard cap inputs, period selector
- Permission: admin.manage_settings on both routes
Tests: 766/766 vitest (was 756) — +10 budget tests covering enforce/
disabled/cap-exceed/estimate-exceed/soft-warn/period boundaries/
cross-port isolation/silent ledger failure.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 19:53:09 +02:00
async function runOpenAi ( { imageBuffer , mimeType , apiKey , model } : RunArgs ) : Promise < OcrRunResult > {
feat(phase-b): ship analytics dashboard, alerts, scanner PWA, dedup, audit view
Phase B (Insights & Alerts) PR4-11 in one drop. Builds on the schema +
service skeletons committed in PRs 1-3.
PR4 Analytics dashboard — 4 chart types (funnel/timeline/breakdown/source),
date-range picker (today/7d/30d/90d), CSV+PNG export per card.
PR5 Alert rail UI + /alerts page — topbar bell w/ live count, dashboard
right-rail, three-tab page (active/dismissed/resolved), socket-driven
invalidation. Bell lazy-loads list on popover open to keep cold pages
fast in non-dashboard routes.
PR6 EOI queue tab on documents hub — filters to in-flight EOIs, count
surfaces in tab label.
PR7 Interests-by-berth tab on berth detail — replaces the stub.
PR8 Expense duplicate detection — BullMQ job runs scan on create, yellow
banner on detail w/ Merge / Not-a-duplicate, transactional merge
consolidates receipts and archives the source.
PR9 Receipt scanner PWA + multi-provider AI — port-scoped /scan route in
its own (scanner) group with no dashboard chrome, dynamic per-port
manifest, OpenAI + Claude provider abstraction, admin OCR settings
page (port-level + super-admin global default w/ opt-in fallback),
test-connection endpoint, manual-entry fallback when no key is
configured. Verify form always shown before save — no ghost rows.
PR10 Audit log read view — swap to tsvector full-text search on the
existing GIN index, cursor pagination, filters for entity/action/user
/date range, batched actor-email resolution.
PR11 Real-API tests — opt-in receipt-ocr.spec (admin save+test, optional
real-receipt parse via REALAPI_RECEIPT_FIXTURE) and alert-engine
socket-fanout spec gated behind RUN_ALERT_ENGINE_REALAPI. Both skip
cleanly without their gate envs so CI stays green.
Test totals: vitest 690 -> 713, smoke 130 -> 138, realapi +2 opt-in.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 17:21:55 +02:00
const client = new OpenAI ( { apiKey } ) ;
const base64 = imageBuffer . toString ( 'base64' ) ;
const response = await client . chat . completions . create ( {
model ,
messages : [
{ role : 'system' , content : SYSTEM_PROMPT } ,
{
role : 'user' ,
content : [
{ type : 'text' , text : 'Extract the receipt as JSON.' } ,
{
type : 'image_url' ,
image_url : { url : ` data: ${ mimeType } ;base64, ${ base64 } ` } ,
} ,
] ,
} ,
] ,
max_tokens : 1024 ,
response_format : { type : 'json_object' } ,
} ) ;
feat(ai): per-port token budgets + usage ledger for AI features
Adds a token-denominated guardrail in front of every server-side AI call
so a misconfigured port can't run up an unbounded bill. Soft caps surface
a banner; hard caps refuse new requests until the period rolls over.
Usage flows into a feature-typed ledger so future AI surfaces (summary,
embeddings, reply-draft) can drop in without schema changes.
- New table ai_usage_ledger (port, user, feature, provider, model,
input/output/total tokens, request id) with two indexes for rollup
- New service ai-budget.service.ts: getAiBudget/setAiBudget,
checkBudget (pre-flight gate), recordAiUsage, currentPeriodTokens,
periodBreakdown — all token-based, period boundaries in UTC
- runOcr now returns provider usage so the route can record the actual
spend instead of estimating
- Scan-receipt route gates on checkBudget before invoking AI; returns
source: manual / reason: budget-exceeded when blocked, surfaces
softCapWarning on the success path
- Admin UI: new AiBudgetCard on the OCR settings page — shows current
spend, per-feature breakdown, soft/hard cap inputs, period selector
- Permission: admin.manage_settings on both routes
Tests: 766/766 vitest (was 756) — +10 budget tests covering enforce/
disabled/cap-exceed/estimate-exceed/soft-warn/period boundaries/
cross-port isolation/silent ledger failure.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 19:53:09 +02:00
const parsed = safeParse ( response . choices [ 0 ] ? . message ? . content ? ? '{}' ) ;
return {
parsed ,
usage : {
inputTokens : response.usage?.prompt_tokens ? ? 0 ,
outputTokens : response.usage?.completion_tokens ? ? 0 ,
requestId : response.id ? ? null ,
} ,
} ;
feat(phase-b): ship analytics dashboard, alerts, scanner PWA, dedup, audit view
Phase B (Insights & Alerts) PR4-11 in one drop. Builds on the schema +
service skeletons committed in PRs 1-3.
PR4 Analytics dashboard — 4 chart types (funnel/timeline/breakdown/source),
date-range picker (today/7d/30d/90d), CSV+PNG export per card.
PR5 Alert rail UI + /alerts page — topbar bell w/ live count, dashboard
right-rail, three-tab page (active/dismissed/resolved), socket-driven
invalidation. Bell lazy-loads list on popover open to keep cold pages
fast in non-dashboard routes.
PR6 EOI queue tab on documents hub — filters to in-flight EOIs, count
surfaces in tab label.
PR7 Interests-by-berth tab on berth detail — replaces the stub.
PR8 Expense duplicate detection — BullMQ job runs scan on create, yellow
banner on detail w/ Merge / Not-a-duplicate, transactional merge
consolidates receipts and archives the source.
PR9 Receipt scanner PWA + multi-provider AI — port-scoped /scan route in
its own (scanner) group with no dashboard chrome, dynamic per-port
manifest, OpenAI + Claude provider abstraction, admin OCR settings
page (port-level + super-admin global default w/ opt-in fallback),
test-connection endpoint, manual-entry fallback when no key is
configured. Verify form always shown before save — no ghost rows.
PR10 Audit log read view — swap to tsvector full-text search on the
existing GIN index, cursor pagination, filters for entity/action/user
/date range, batched actor-email resolution.
PR11 Real-API tests — opt-in receipt-ocr.spec (admin save+test, optional
real-receipt parse via REALAPI_RECEIPT_FIXTURE) and alert-engine
socket-fanout spec gated behind RUN_ALERT_ENGINE_REALAPI. Both skip
cleanly without their gate envs so CI stays green.
Test totals: vitest 690 -> 713, smoke 130 -> 138, realapi +2 opt-in.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 17:21:55 +02:00
}
feat(ai): per-port token budgets + usage ledger for AI features
Adds a token-denominated guardrail in front of every server-side AI call
so a misconfigured port can't run up an unbounded bill. Soft caps surface
a banner; hard caps refuse new requests until the period rolls over.
Usage flows into a feature-typed ledger so future AI surfaces (summary,
embeddings, reply-draft) can drop in without schema changes.
- New table ai_usage_ledger (port, user, feature, provider, model,
input/output/total tokens, request id) with two indexes for rollup
- New service ai-budget.service.ts: getAiBudget/setAiBudget,
checkBudget (pre-flight gate), recordAiUsage, currentPeriodTokens,
periodBreakdown — all token-based, period boundaries in UTC
- runOcr now returns provider usage so the route can record the actual
spend instead of estimating
- Scan-receipt route gates on checkBudget before invoking AI; returns
source: manual / reason: budget-exceeded when blocked, surfaces
softCapWarning on the success path
- Admin UI: new AiBudgetCard on the OCR settings page — shows current
spend, per-feature breakdown, soft/hard cap inputs, period selector
- Permission: admin.manage_settings on both routes
Tests: 766/766 vitest (was 756) — +10 budget tests covering enforce/
disabled/cap-exceed/estimate-exceed/soft-warn/period boundaries/
cross-port isolation/silent ledger failure.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 19:53:09 +02:00
async function runClaude ( { imageBuffer , mimeType , apiKey , model } : RunArgs ) : Promise < OcrRunResult > {
feat(phase-b): ship analytics dashboard, alerts, scanner PWA, dedup, audit view
Phase B (Insights & Alerts) PR4-11 in one drop. Builds on the schema +
service skeletons committed in PRs 1-3.
PR4 Analytics dashboard — 4 chart types (funnel/timeline/breakdown/source),
date-range picker (today/7d/30d/90d), CSV+PNG export per card.
PR5 Alert rail UI + /alerts page — topbar bell w/ live count, dashboard
right-rail, three-tab page (active/dismissed/resolved), socket-driven
invalidation. Bell lazy-loads list on popover open to keep cold pages
fast in non-dashboard routes.
PR6 EOI queue tab on documents hub — filters to in-flight EOIs, count
surfaces in tab label.
PR7 Interests-by-berth tab on berth detail — replaces the stub.
PR8 Expense duplicate detection — BullMQ job runs scan on create, yellow
banner on detail w/ Merge / Not-a-duplicate, transactional merge
consolidates receipts and archives the source.
PR9 Receipt scanner PWA + multi-provider AI — port-scoped /scan route in
its own (scanner) group with no dashboard chrome, dynamic per-port
manifest, OpenAI + Claude provider abstraction, admin OCR settings
page (port-level + super-admin global default w/ opt-in fallback),
test-connection endpoint, manual-entry fallback when no key is
configured. Verify form always shown before save — no ghost rows.
PR10 Audit log read view — swap to tsvector full-text search on the
existing GIN index, cursor pagination, filters for entity/action/user
/date range, batched actor-email resolution.
PR11 Real-API tests — opt-in receipt-ocr.spec (admin save+test, optional
real-receipt parse via REALAPI_RECEIPT_FIXTURE) and alert-engine
socket-fanout spec gated behind RUN_ALERT_ENGINE_REALAPI. Both skip
cleanly without their gate envs so CI stays green.
Test totals: vitest 690 -> 713, smoke 130 -> 138, realapi +2 opt-in.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 17:21:55 +02:00
const base64 = imageBuffer . toString ( 'base64' ) ;
const res = await fetch ( 'https://api.anthropic.com/v1/messages' , {
method : 'POST' ,
headers : {
'Content-Type' : 'application/json' ,
'x-api-key' : apiKey ,
'anthropic-version' : '2023-06-01' ,
} ,
body : JSON.stringify ( {
model ,
max_tokens : 1024 ,
system : SYSTEM_PROMPT ,
messages : [
{
role : 'user' ,
content : [
{
type : 'image' ,
source : { type : 'base64' , media_type : mimeType , data : base64 } ,
} ,
{ type : 'text' , text : 'Extract the receipt as JSON.' } ,
] ,
} ,
] ,
} ) ,
} ) ;
if ( ! res . ok ) {
const detail = await res . text ( ) . catch ( ( ) = > '' ) ;
throw new Error ( ` Claude API ${ res . status } : ${ detail . slice ( 0 , 200 ) } ` ) ;
}
feat(ai): per-port token budgets + usage ledger for AI features
Adds a token-denominated guardrail in front of every server-side AI call
so a misconfigured port can't run up an unbounded bill. Soft caps surface
a banner; hard caps refuse new requests until the period rolls over.
Usage flows into a feature-typed ledger so future AI surfaces (summary,
embeddings, reply-draft) can drop in without schema changes.
- New table ai_usage_ledger (port, user, feature, provider, model,
input/output/total tokens, request id) with two indexes for rollup
- New service ai-budget.service.ts: getAiBudget/setAiBudget,
checkBudget (pre-flight gate), recordAiUsage, currentPeriodTokens,
periodBreakdown — all token-based, period boundaries in UTC
- runOcr now returns provider usage so the route can record the actual
spend instead of estimating
- Scan-receipt route gates on checkBudget before invoking AI; returns
source: manual / reason: budget-exceeded when blocked, surfaces
softCapWarning on the success path
- Admin UI: new AiBudgetCard on the OCR settings page — shows current
spend, per-feature breakdown, soft/hard cap inputs, period selector
- Permission: admin.manage_settings on both routes
Tests: 766/766 vitest (was 756) — +10 budget tests covering enforce/
disabled/cap-exceed/estimate-exceed/soft-warn/period boundaries/
cross-port isolation/silent ledger failure.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 19:53:09 +02:00
const body = ( await res . json ( ) ) as {
id? : string ;
content? : Array < { type : string ; text? : string } > ;
usage ? : { input_tokens? : number ; output_tokens? : number } ;
} ;
feat(phase-b): ship analytics dashboard, alerts, scanner PWA, dedup, audit view
Phase B (Insights & Alerts) PR4-11 in one drop. Builds on the schema +
service skeletons committed in PRs 1-3.
PR4 Analytics dashboard — 4 chart types (funnel/timeline/breakdown/source),
date-range picker (today/7d/30d/90d), CSV+PNG export per card.
PR5 Alert rail UI + /alerts page — topbar bell w/ live count, dashboard
right-rail, three-tab page (active/dismissed/resolved), socket-driven
invalidation. Bell lazy-loads list on popover open to keep cold pages
fast in non-dashboard routes.
PR6 EOI queue tab on documents hub — filters to in-flight EOIs, count
surfaces in tab label.
PR7 Interests-by-berth tab on berth detail — replaces the stub.
PR8 Expense duplicate detection — BullMQ job runs scan on create, yellow
banner on detail w/ Merge / Not-a-duplicate, transactional merge
consolidates receipts and archives the source.
PR9 Receipt scanner PWA + multi-provider AI — port-scoped /scan route in
its own (scanner) group with no dashboard chrome, dynamic per-port
manifest, OpenAI + Claude provider abstraction, admin OCR settings
page (port-level + super-admin global default w/ opt-in fallback),
test-connection endpoint, manual-entry fallback when no key is
configured. Verify form always shown before save — no ghost rows.
PR10 Audit log read view — swap to tsvector full-text search on the
existing GIN index, cursor pagination, filters for entity/action/user
/date range, batched actor-email resolution.
PR11 Real-API tests — opt-in receipt-ocr.spec (admin save+test, optional
real-receipt parse via REALAPI_RECEIPT_FIXTURE) and alert-engine
socket-fanout spec gated behind RUN_ALERT_ENGINE_REALAPI. Both skip
cleanly without their gate envs so CI stays green.
Test totals: vitest 690 -> 713, smoke 130 -> 138, realapi +2 opt-in.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 17:21:55 +02:00
const text = body . content ? . find ( ( c ) = > c . type === 'text' ) ? . text ? ? '{}' ;
feat(ai): per-port token budgets + usage ledger for AI features
Adds a token-denominated guardrail in front of every server-side AI call
so a misconfigured port can't run up an unbounded bill. Soft caps surface
a banner; hard caps refuse new requests until the period rolls over.
Usage flows into a feature-typed ledger so future AI surfaces (summary,
embeddings, reply-draft) can drop in without schema changes.
- New table ai_usage_ledger (port, user, feature, provider, model,
input/output/total tokens, request id) with two indexes for rollup
- New service ai-budget.service.ts: getAiBudget/setAiBudget,
checkBudget (pre-flight gate), recordAiUsage, currentPeriodTokens,
periodBreakdown — all token-based, period boundaries in UTC
- runOcr now returns provider usage so the route can record the actual
spend instead of estimating
- Scan-receipt route gates on checkBudget before invoking AI; returns
source: manual / reason: budget-exceeded when blocked, surfaces
softCapWarning on the success path
- Admin UI: new AiBudgetCard on the OCR settings page — shows current
spend, per-feature breakdown, soft/hard cap inputs, period selector
- Permission: admin.manage_settings on both routes
Tests: 766/766 vitest (was 756) — +10 budget tests covering enforce/
disabled/cap-exceed/estimate-exceed/soft-warn/period boundaries/
cross-port isolation/silent ledger failure.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 19:53:09 +02:00
const parsed = safeParse ( text ) ;
return {
parsed ,
usage : {
inputTokens : body.usage?.input_tokens ? ? 0 ,
outputTokens : body.usage?.output_tokens ? ? 0 ,
requestId : body.id ? ? null ,
} ,
} ;
feat(phase-b): ship analytics dashboard, alerts, scanner PWA, dedup, audit view
Phase B (Insights & Alerts) PR4-11 in one drop. Builds on the schema +
service skeletons committed in PRs 1-3.
PR4 Analytics dashboard — 4 chart types (funnel/timeline/breakdown/source),
date-range picker (today/7d/30d/90d), CSV+PNG export per card.
PR5 Alert rail UI + /alerts page — topbar bell w/ live count, dashboard
right-rail, three-tab page (active/dismissed/resolved), socket-driven
invalidation. Bell lazy-loads list on popover open to keep cold pages
fast in non-dashboard routes.
PR6 EOI queue tab on documents hub — filters to in-flight EOIs, count
surfaces in tab label.
PR7 Interests-by-berth tab on berth detail — replaces the stub.
PR8 Expense duplicate detection — BullMQ job runs scan on create, yellow
banner on detail w/ Merge / Not-a-duplicate, transactional merge
consolidates receipts and archives the source.
PR9 Receipt scanner PWA + multi-provider AI — port-scoped /scan route in
its own (scanner) group with no dashboard chrome, dynamic per-port
manifest, OpenAI + Claude provider abstraction, admin OCR settings
page (port-level + super-admin global default w/ opt-in fallback),
test-connection endpoint, manual-entry fallback when no key is
configured. Verify form always shown before save — no ghost rows.
PR10 Audit log read view — swap to tsvector full-text search on the
existing GIN index, cursor pagination, filters for entity/action/user
/date range, batched actor-email resolution.
PR11 Real-API tests — opt-in receipt-ocr.spec (admin save+test, optional
real-receipt parse via REALAPI_RECEIPT_FIXTURE) and alert-engine
socket-fanout spec gated behind RUN_ALERT_ENGINE_REALAPI. Both skip
cleanly without their gate envs so CI stays green.
Test totals: vitest 690 -> 713, smoke 130 -> 138, realapi +2 opt-in.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 17:21:55 +02:00
}
export async function runOcr ( args : {
provider : 'openai' | 'claude' ;
imageBuffer : Buffer ;
mimeType : string ;
apiKey : string ;
model : string ;
feat(ai): per-port token budgets + usage ledger for AI features
Adds a token-denominated guardrail in front of every server-side AI call
so a misconfigured port can't run up an unbounded bill. Soft caps surface
a banner; hard caps refuse new requests until the period rolls over.
Usage flows into a feature-typed ledger so future AI surfaces (summary,
embeddings, reply-draft) can drop in without schema changes.
- New table ai_usage_ledger (port, user, feature, provider, model,
input/output/total tokens, request id) with two indexes for rollup
- New service ai-budget.service.ts: getAiBudget/setAiBudget,
checkBudget (pre-flight gate), recordAiUsage, currentPeriodTokens,
periodBreakdown — all token-based, period boundaries in UTC
- runOcr now returns provider usage so the route can record the actual
spend instead of estimating
- Scan-receipt route gates on checkBudget before invoking AI; returns
source: manual / reason: budget-exceeded when blocked, surfaces
softCapWarning on the success path
- Admin UI: new AiBudgetCard on the OCR settings page — shows current
spend, per-feature breakdown, soft/hard cap inputs, period selector
- Permission: admin.manage_settings on both routes
Tests: 766/766 vitest (was 756) — +10 budget tests covering enforce/
disabled/cap-exceed/estimate-exceed/soft-warn/period boundaries/
cross-port isolation/silent ledger failure.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 19:53:09 +02:00
} ) : Promise < OcrRunResult > {
feat(phase-b): ship analytics dashboard, alerts, scanner PWA, dedup, audit view
Phase B (Insights & Alerts) PR4-11 in one drop. Builds on the schema +
service skeletons committed in PRs 1-3.
PR4 Analytics dashboard — 4 chart types (funnel/timeline/breakdown/source),
date-range picker (today/7d/30d/90d), CSV+PNG export per card.
PR5 Alert rail UI + /alerts page — topbar bell w/ live count, dashboard
right-rail, three-tab page (active/dismissed/resolved), socket-driven
invalidation. Bell lazy-loads list on popover open to keep cold pages
fast in non-dashboard routes.
PR6 EOI queue tab on documents hub — filters to in-flight EOIs, count
surfaces in tab label.
PR7 Interests-by-berth tab on berth detail — replaces the stub.
PR8 Expense duplicate detection — BullMQ job runs scan on create, yellow
banner on detail w/ Merge / Not-a-duplicate, transactional merge
consolidates receipts and archives the source.
PR9 Receipt scanner PWA + multi-provider AI — port-scoped /scan route in
its own (scanner) group with no dashboard chrome, dynamic per-port
manifest, OpenAI + Claude provider abstraction, admin OCR settings
page (port-level + super-admin global default w/ opt-in fallback),
test-connection endpoint, manual-entry fallback when no key is
configured. Verify form always shown before save — no ghost rows.
PR10 Audit log read view — swap to tsvector full-text search on the
existing GIN index, cursor pagination, filters for entity/action/user
/date range, batched actor-email resolution.
PR11 Real-API tests — opt-in receipt-ocr.spec (admin save+test, optional
real-receipt parse via REALAPI_RECEIPT_FIXTURE) and alert-engine
socket-fanout spec gated behind RUN_ALERT_ENGINE_REALAPI. Both skip
cleanly without their gate envs so CI stays green.
Test totals: vitest 690 -> 713, smoke 130 -> 138, realapi +2 opt-in.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 17:21:55 +02:00
if ( args . provider === 'openai' ) return runOpenAi ( args ) ;
return runClaude ( args ) ;
}
/ * *
* Tiny dummy - image probe used by the admin "Test connection" button .
* Returns the raw HTTP status so callers can render plain - English errors .
* /
export async function testProvider (
provider : 'openai' | 'claude' ,
apiKey : string ,
model : string ,
) : Promise < { ok : true } | { ok : false ; reason : string } > {
// 1× 1 transparent PNG.
const pixelPng = Buffer . from (
'iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNk+A8AAQUBAScY42YAAAAASUVORK5CYII=' ,
'base64' ,
) ;
try {
await runOcr ( {
provider ,
imageBuffer : pixelPng ,
mimeType : 'image/png' ,
apiKey ,
model ,
} ) ;
return { ok : true } ;
} catch ( err ) {
const reason = err instanceof Error ? err . message : 'Unknown error' ;
return { ok : false , reason } ;
}
}
feat(ai): per-port token budgets + usage ledger for AI features
Adds a token-denominated guardrail in front of every server-side AI call
so a misconfigured port can't run up an unbounded bill. Soft caps surface
a banner; hard caps refuse new requests until the period rolls over.
Usage flows into a feature-typed ledger so future AI surfaces (summary,
embeddings, reply-draft) can drop in without schema changes.
- New table ai_usage_ledger (port, user, feature, provider, model,
input/output/total tokens, request id) with two indexes for rollup
- New service ai-budget.service.ts: getAiBudget/setAiBudget,
checkBudget (pre-flight gate), recordAiUsage, currentPeriodTokens,
periodBreakdown — all token-based, period boundaries in UTC
- runOcr now returns provider usage so the route can record the actual
spend instead of estimating
- Scan-receipt route gates on checkBudget before invoking AI; returns
source: manual / reason: budget-exceeded when blocked, surfaces
softCapWarning on the success path
- Admin UI: new AiBudgetCard on the OCR settings page — shows current
spend, per-feature breakdown, soft/hard cap inputs, period selector
- Permission: admin.manage_settings on both routes
Tests: 766/766 vitest (was 756) — +10 budget tests covering enforce/
disabled/cap-exceed/estimate-exceed/soft-warn/period boundaries/
cross-port isolation/silent ledger failure.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-28 19:53:09 +02:00
export const OCR_FEATURE = 'ocr_receipt' ;
export const OCR_ESTIMATED_TOKENS = 2048 ;