@exulu/backend 3.5.0 → 3.6.0

package/dist/index.cjs

@@ -11452,6 +11452,52 @@ var usersSchema = {
       // triggered by the key with project_id_ for LiteLLM cost attribution.
       name: "project",
       type: "uuid"
+    },
+    // ── Lead capture (see lead-capture spec §5) ─────────────────────────────
+    // Consent must be *provable* under Art. 7(1) GDPR: who agreed, when, and
+    // to which wording. Hence a timestamp per consent and a version string
+    // pointing at the exact text that was shown.
+    {
+      // Current state of the optional marketing consent.
+      name: "marketing_consent",
+      type: "boolean",
+      default: false
+    },
+    {
+      // When marketing consent was GIVEN.
+      name: "marketing_consent_at",
+      type: "date"
+    },
+    {
+      // When marketing consent was WITHDRAWN. Deliberately a separate column:
+      // overwriting marketing_consent_at on withdrawal would destroy the grant
+      // time, and with it the ability to prove the order of events if someone
+      // later claims they were mailed after withdrawing.
+      name: "marketing_consent_withdrawn_at",
+      type: "date"
+    },
+    {
+      // When the mandatory storage/processing consent was given.
+      name: "processing_consent_at",
+      type: "date"
+    },
+    {
+      // When the user asked for erasure (Art. 17). The row survives so the
+      // one-month deadline stays queryable even if the notification mail fails.
+      name: "erasure_requested_at",
+      type: "date"
+    },
+    {
+      // Where the lead came from, e.g. "eu-ai-act-bot".
+      name: "signup_source",
+      type: "text"
+    },
+    {
+      // Version of the consent wording that was actually displayed, e.g.
+      // "2026-08-18". The wording itself lives in the frontend repo under
+      // version control; this column points at it.
+      name: "consent_version",
+      type: "text"
     }
   ]
 };
@@ -11760,6 +11806,32 @@ var addCoreFields = (schema) => {
   }
   return schema;
 };
+var otpSendAttemptsSchema = {
+  type: "otp_send_attempts",
+  name: {
+    plural: "otp_send_attempts",
+    singular: "otp_send_attempt"
+  },
+  fields: [
+    {
+      name: "key",
+      type: "text",
+      index: true,
+      unique: true,
+      required: true
+    },
+    {
+      name: "count",
+      type: "number",
+      required: true
+    },
+    {
+      name: "window_start",
+      type: "date",
+      required: true
+    }
+  ]
+};
 var coreSchemas = {
   get: () => {
     const license = checkLicense();
@@ -11780,7 +11852,8 @@ var coreSchemas = {
       contextPresetsSchema: () => addCoreFields(contextPresetsSchema),
       sharedArtifactsSchema: () => addCoreFields(sharedArtifactsSchema),
       transcriptionJobsSchema: () => addCoreFields(transcriptionJobsSchema),
-      imageGenerationsSchema: () => addCoreFields(imageGenerationsSchema)
+      imageGenerationsSchema: () => addCoreFields(imageGenerationsSchema),
+      otpSendAttemptsSchema: () => addCoreFields(otpSendAttemptsSchema)
     };
     if (license["agent-feedback"]) {
       schemas.feedbackSchema = () => addCoreFields(feedbackSchema);
@@ -31560,7 +31633,8 @@ var {
   promptFavoritesSchema: promptFavoritesSchema3,
   transcriptionJobsSchema: transcriptionJobsSchema3,
   imageGenerationsSchema: imageGenerationsSchema2,
-  sharedArtifactsSchema: sharedArtifactsSchema2
+  sharedArtifactsSchema: sharedArtifactsSchema2,
+  otpSendAttemptsSchema: otpSendAttemptsSchema2
 } = coreSchemas.get();
 var addMissingFields = async (knex, tableName, fields, skipFields = []) => {
   for (const field of fields) {
@@ -31631,7 +31705,8 @@ var up = async function(knex) {
     variablesSchema3(),
     skillsSchema3(),
     workflowTemplatesSchema3(),
-    workflowTriggersSchema2()
+    workflowTriggersSchema2(),
+    otpSendAttemptsSchema2()
   ];
   const createTable = async (schema) => {
     if (!await knex.schema.hasTable(schema.name.plural)) {
@@ -33398,7 +33473,16 @@ ${setupResult.output || ""}`);
         args: [paths.source, chunksDir, "--chunk-size", String(maxPagesPerChunk), "--max-size-mb", "25"],
         timeout: 5 * 60 * 1e3
       });
-      const pdfChunks = JSON.parse(splitResult.stdout);
+      let pdfChunks;
+      try {
+        pdfChunks = JSON.parse(splitResult.stdout);
+      } catch (err) {
+        throw new Error(
+          `[EXULU] split_pdf.py returned invalid JSON on stdout: ${err.message}
+stdout: ${splitResult.stdout.slice(0, 500)}
+stderr: ${splitResult.stderr.slice(-1e3)}`
+        );
+      }
       console.log(`[EXULU] PDF split into ${pdfChunks.length} chunk(s) for OCR (max ${maxPagesPerChunk} pages each)`);
       const chunkLimit = (0, import_p_limit.default)(3);
       const chunkResults = await Promise.all(

package/dist/index.js

@@ -2568,6 +2568,52 @@ var usersSchema = {
       // triggered by the key with project_id_ for LiteLLM cost attribution.
       name: "project",
       type: "uuid"
+    },
+    // ── Lead capture (see lead-capture spec §5) ─────────────────────────────
+    // Consent must be *provable* under Art. 7(1) GDPR: who agreed, when, and
+    // to which wording. Hence a timestamp per consent and a version string
+    // pointing at the exact text that was shown.
+    {
+      // Current state of the optional marketing consent.
+      name: "marketing_consent",
+      type: "boolean",
+      default: false
+    },
+    {
+      // When marketing consent was GIVEN.
+      name: "marketing_consent_at",
+      type: "date"
+    },
+    {
+      // When marketing consent was WITHDRAWN. Deliberately a separate column:
+      // overwriting marketing_consent_at on withdrawal would destroy the grant
+      // time, and with it the ability to prove the order of events if someone
+      // later claims they were mailed after withdrawing.
+      name: "marketing_consent_withdrawn_at",
+      type: "date"
+    },
+    {
+      // When the mandatory storage/processing consent was given.
+      name: "processing_consent_at",
+      type: "date"
+    },
+    {
+      // When the user asked for erasure (Art. 17). The row survives so the
+      // one-month deadline stays queryable even if the notification mail fails.
+      name: "erasure_requested_at",
+      type: "date"
+    },
+    {
+      // Where the lead came from, e.g. "eu-ai-act-bot".
+      name: "signup_source",
+      type: "text"
+    },
+    {
+      // Version of the consent wording that was actually displayed, e.g.
+      // "2026-08-18". The wording itself lives in the frontend repo under
+      // version control; this column points at it.
+      name: "consent_version",
+      type: "text"
     }
   ]
 };
@@ -2876,6 +2922,32 @@ var addCoreFields = (schema) => {
   }
   return schema;
 };
+var otpSendAttemptsSchema = {
+  type: "otp_send_attempts",
+  name: {
+    plural: "otp_send_attempts",
+    singular: "otp_send_attempt"
+  },
+  fields: [
+    {
+      name: "key",
+      type: "text",
+      index: true,
+      unique: true,
+      required: true
+    },
+    {
+      name: "count",
+      type: "number",
+      required: true
+    },
+    {
+      name: "window_start",
+      type: "date",
+      required: true
+    }
+  ]
+};
 var coreSchemas = {
   get: () => {
     const license = checkLicense();
@@ -2896,7 +2968,8 @@ var coreSchemas = {
       contextPresetsSchema: () => addCoreFields(contextPresetsSchema),
       sharedArtifactsSchema: () => addCoreFields(sharedArtifactsSchema),
       transcriptionJobsSchema: () => addCoreFields(transcriptionJobsSchema),
-      imageGenerationsSchema: () => addCoreFields(imageGenerationsSchema)
+      imageGenerationsSchema: () => addCoreFields(imageGenerationsSchema),
+      otpSendAttemptsSchema: () => addCoreFields(otpSendAttemptsSchema)
     };
     if (license["agent-feedback"]) {
       schemas.feedbackSchema = () => addCoreFields(feedbackSchema);
@@ -22174,7 +22247,8 @@ var {
   promptFavoritesSchema: promptFavoritesSchema3,
   transcriptionJobsSchema: transcriptionJobsSchema3,
   imageGenerationsSchema: imageGenerationsSchema2,
-  sharedArtifactsSchema: sharedArtifactsSchema2
+  sharedArtifactsSchema: sharedArtifactsSchema2,
+  otpSendAttemptsSchema: otpSendAttemptsSchema2
 } = coreSchemas.get();
 var addMissingFields = async (knex, tableName, fields, skipFields = []) => {
   for (const field of fields) {
@@ -22245,7 +22319,8 @@ var up = async function(knex) {
     variablesSchema3(),
     skillsSchema3(),
     workflowTemplatesSchema3(),
-    workflowTriggersSchema2()
+    workflowTriggersSchema2(),
+    otpSendAttemptsSchema2()
   ];
   const createTable = async (schema) => {
     if (!await knex.schema.hasTable(schema.name.plural)) {
@@ -23992,7 +24067,16 @@ ${setupResult.output || ""}`);
         args: [paths.source, chunksDir, "--chunk-size", String(maxPagesPerChunk), "--max-size-mb", "25"],
         timeout: 5 * 60 * 1e3
       });
-      const pdfChunks = JSON.parse(splitResult.stdout);
+      let pdfChunks;
+      try {
+        pdfChunks = JSON.parse(splitResult.stdout);
+      } catch (err) {
+        throw new Error(
+          `[EXULU] split_pdf.py returned invalid JSON on stdout: ${err.message}
+stdout: ${splitResult.stdout.slice(0, 500)}
+stderr: ${splitResult.stderr.slice(-1e3)}`
+        );
+      }
       console.log(`[EXULU] PDF split into ${pdfChunks.length} chunk(s) for OCR (max ${maxPagesPerChunk} pages each)`);
       const chunkLimit = pLimit(3);
       const chunkResults = await Promise.all(

package/ee/python/documents/processing/doc_processor.ts

@@ -837,8 +837,20 @@ async function processPdf(
         timeout: 5 * 60 * 1000,
       });
 
-      const pdfChunks: Array<{ path: string; start_page: number; end_page: number }> =
-        JSON.parse(splitResult.stdout);
+      // split_pdf.py contracts to put nothing but the JSON payload on stdout.
+      // If something slips in anyway, the bare SyntaxError only quotes the first
+      // few characters ("Unexpected token 'w'") and names neither the script nor
+      // the offending output — so re-throw with the actual streams attached.
+      let pdfChunks: Array<{ path: string; start_page: number; end_page: number }>;
+      try {
+        pdfChunks = JSON.parse(splitResult.stdout);
+      } catch (err) {
+        throw new Error(
+          `[EXULU] split_pdf.py returned invalid JSON on stdout: ${(err as Error).message}\n` +
+          `stdout: ${splitResult.stdout.slice(0, 500)}\n` +
+          `stderr: ${splitResult.stderr.slice(-1000)}`
+        );
+      }
 
       console.log(`[EXULU] PDF split into ${pdfChunks.length} chunk(s) for OCR (max ${maxPagesPerChunk} pages each)`);
 

package/ee/python/documents/processing/split_pdf.py

@@ -20,7 +20,24 @@ import os
 import json
 import argparse
 
-import fitz  # PyMuPDF — installed as a docling transitive dependency
+# stdout is this script's result channel and the caller does JSON.parse() on it,
+# so nothing else may write there. Our own prints all pass file=sys.stderr, but
+# dependencies do not honour that: PyMuPDF sends its messages to sys.stdout by
+# default, and importing the legacy `fitz` alias emits
+# "warning: The `fitz` API is deprecated ..." — which lands ahead of the payload
+# and fails the caller with "Unexpected token 'w'". PyMuPDF arrives unpinned as
+# a docling transitive dependency, so a routine rebuild is enough to introduce a
+# banner like that. Point sys.stdout at stderr before importing anything and
+# keep a private handle for the result, so any library that prints — now or
+# after a future dependency bump — is shunted to the log channel instead of
+# corrupting the payload.
+_stdout = sys.stdout
+sys.stdout = sys.stderr
+
+try:
+    import pymupdf as fitz  # PyMuPDF >= 1.24.3, where the module was renamed
+except ImportError:  # older releases only ship the legacy `fitz` module
+    import fitz
 
 
 def _write_chunk(
@@ -145,7 +162,7 @@ if __name__ == "__main__":
 
     try:
         chunks = split_pdf(args.input_pdf, args.output_dir, args.chunk_size, max_size_bytes)
-        print(json.dumps(chunks))
+        print(json.dumps(chunks), file=_stdout)
     except Exception as e:
         print(f"[split_pdf] ERROR: {e}", file=sys.stderr)
         sys.exit(1)

package/ee/python/requirements.txt

@@ -24,17 +24,26 @@ pyannote.audio>=3.3.0
 # Belt-and-suspenders: keep huggingface_hub on the 0.x line so pyannote 3.x's
 # `use_auth_token=` calls keep working (1.x removed that kwarg → diarization off).
 huggingface_hub<1.0
-fastapi
 uvicorn
 python-multipart
 requests
 # LiteLLM proxy — only used when EXULU_USE_LITELLM=true. Always installed so
 # the dep is ready when the env var is flipped. Pinned to a tested version;
 # upgrade deliberately.
-litellm[proxy]==1.85.1
+litellm[proxy]==1.97.0
+# FastAPI ceiling — NOT cosmetic. litellm 1.97.0 declares fastapi>=0.136.3,<1.0
+# but imports the private `get_flat_dependant` from fastapi.dependencies.utils
+# (proxy/management_endpoints/management_v1/common.py:7), which FastAPI deleted
+# in 0.140.7. Any resolve that picks 0.140.7+ yields a proxy that cannot start:
+# "ImportError: cannot import name 'get_flat_dependant'". 0.140.6 is the last
+# version that still has it. NOTE: fastapi is also a first-party dependency of
+# ee/python/transcription/server.py (not just a litellm transitive). Once
+# litellm ships its own ceiling, replace this constrained pin with a bare
+# `fastapi` line — do not delete it entirely.
+fastapi>=0.136.3,<0.140.7  # to confirm this pin is still needed: grep -r get_flat_dependant $(pip show litellm | awk '/Location/{print $2}')/litellm/proxy/
 # Prisma Python client — required by LiteLLM proxy whenever a `database_url`
 # is set in config.litellm.yaml (used for user/budget tracking, key
-# management, etc.). NOT included in litellm[proxy] in 1.85.1, hence the
+# management, etc.). NOT included in litellm[proxy] in 1.97.0, hence the
 # separate pin. setup.sh runs `prisma generate` against LiteLLM's bundled
 # schema after pip install so the Python client module is materialized.
 prisma==0.15.0

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@exulu/backend",
   "author": "Qventu Bv.",
-  "version": "3.5.0",
+  "version": "3.6.0",
   "main": "./dist/index.js",
   "private": false,
   "publishConfig": {