ux cleanup

modules/auth/trustedDeviceService.py

@@ -0,0 +1,171 @@
+# Copyright (c) 2026 PowerOn AG
+# All rights reserved.
+"""
+Trusted Device Service.
+
+After successful MFA verification a device can be marked as trusted for a
+configurable duration (default 60 days). On subsequent logins from the same
+device the MFA step is skipped.
+
+Cookie: ``mfa_trusted`` (httpOnly, Secure, SameSite policy from jwtService).
+DB: ``TrustedDevice`` table in poweron_app.
+
+Regulatory basis:
+- NIST SP 800-63B Section 5.2.8: Verifier MAY re-authenticate only after a
+  configurable period when a device is bound to the subscriber.
+- Microsoft, Google, AWS implement identical patterns.
+"""
+
+import logging
+import secrets
+from typing import Optional
+
+from fastapi import Request, Response
+
+from modules.shared.configuration import APP_CONFIG
+from modules.shared.timeUtils import getUtcNow, getUtcTimestamp
+from modules.datamodels.datamodelSecurity import TrustedDevice
+
+logger = logging.getLogger(__name__)
+
+_COOKIE_NAME = "mfa_trusted"
+_DEFAULT_TRUST_DAYS = 60
+_TOKEN_BYTES = 32
+
+
+def _getTrustDurationDays() -> int:
+    raw = (APP_CONFIG.get("MFA_TRUST_DURATION_DAYS") or "").strip()
+    if raw.isdigit() and int(raw) > 0:
+        return int(raw)
+    return _DEFAULT_TRUST_DAYS
+
+
+def createTrustedDevice(userId: str, request: Request, response: Response, db) -> str:
+    """Create a TrustedDevice entry and set the cookie on the response.
+
+    Returns the device token (cookie value).
+    """
+    from modules.auth.jwtService import _cookiePolicy
+
+    trustDays = _getTrustDurationDays()
+    deviceToken = secrets.token_urlsafe(_TOKEN_BYTES)
+
+    now = getUtcTimestamp()
+    trustedUntil = now + (trustDays * 86400)
+
+    device = TrustedDevice(
+        id=deviceToken,
+        userId=userId,
+        trustedUntil=trustedUntil,
+        userAgent=(request.headers.get("user-agent") or "")[:512],
+        ipAddress=_getClientIp(request),
+        createdAt=now,
+    )
+
+    try:
+        db.recordCreate(TrustedDevice, device.model_dump())
+    except Exception as e:
+        logger.error(f"Failed to persist TrustedDevice for userId={userId}: {e}")
+        return ""
+
+    useSecure, samesite, _ = _cookiePolicy()
+    response.set_cookie(
+        key=_COOKIE_NAME,
+        value=deviceToken,
+        httponly=True,
+        secure=useSecure,
+        samesite=samesite,
+        path="/",
+        max_age=trustDays * 86400,
+    )
+
+    logger.info(f"Trusted device created for userId={userId}, valid {trustDays}d")
+    return deviceToken
+
+
+def isTrustedDevice(request: Request, userId: str, db) -> bool:
+    """Check if the current request comes from a trusted device for the given user."""
+    deviceToken = request.cookies.get(_COOKIE_NAME)
+    if not deviceToken:
+        return False
+
+    try:
+        records = db.getRecordset(
+            TrustedDevice,
+            recordFilter={"id": deviceToken, "userId": userId},
+        )
+        if not records:
+            return False
+
+        device = records[0]
+        trustedUntil = device.get("trustedUntil", 0)
+        if isinstance(trustedUntil, (int, float)) and trustedUntil > getUtcTimestamp():
+            return True
+
+        return False
+    except Exception as e:
+        logger.warning(f"Error checking trusted device for userId={userId}: {e}")
+        return False
+
+
+def revokeTrustedDevices(userId: str, db) -> int:
+    """Revoke all trusted devices for a user. Returns count of deleted entries."""
+    try:
+        records = db.getRecordset(TrustedDevice, recordFilter={"userId": userId})
+        count = 0
+        for rec in records:
+            db.recordDelete(TrustedDevice, rec["id"])
+            count += 1
+        if count:
+            logger.info(f"Revoked {count} trusted device(s) for userId={userId}")
+        return count
+    except Exception as e:
+        logger.error(f"Failed to revoke trusted devices for userId={userId}: {e}")
+        return 0
+
+
+def clearTrustedDeviceCookie(response: Response) -> None:
+    """Clear the mfa_trusted cookie."""
+    from modules.auth.jwtService import _cookiePolicy
+
+    useSecure, samesite, samesiteHeader = _cookiePolicy()
+    secure_flag = "; Secure" if useSecure else ""
+    response.headers.append(
+        "Set-Cookie",
+        f"{_COOKIE_NAME}=deleted; Path=/; Max-Age=0; Expires=Thu, 01 Jan 1970 00:00:00 GMT; HttpOnly{secure_flag}; SameSite={samesiteHeader}"
+    )
+    response.delete_cookie(
+        key=_COOKIE_NAME,
+        path="/",
+        secure=useSecure,
+        httponly=True,
+        samesite=samesite,
+    )
+
+
+def cleanupExpiredDevices(db) -> int:
+    """Remove TrustedDevice entries past their trustedUntil. Returns deleted count."""
+    try:
+        records = db.getRecordset(TrustedDevice, recordFilter={})
+        now = getUtcTimestamp()
+        count = 0
+        for rec in records:
+            if rec.get("trustedUntil", 0) < now:
+                db.recordDelete(TrustedDevice, rec["id"])
+                count += 1
+        if count:
+            logger.info(f"Cleaned up {count} expired trusted device(s)")
+        return count
+    except Exception as e:
+        logger.error(f"Error cleaning up expired trusted devices: {e}")
+        return 0
+
+
+def _getClientIp(request: Request) -> Optional[str]:
+    """Extract client IP from request (respects X-Forwarded-For)."""
+    forwarded = request.headers.get("x-forwarded-for")
+    if forwarded:
+        return forwarded.split(",")[0].strip()
+    if request.client:
+        return request.client.host
+    return None

modules/connectors/connectorDbPostgre.py

@@ -871,6 +871,7 @@ class DatabaseConnector:
                                 ("jsonb", "TEXT"): "TEXT USING \"{col}\"::text",
                                 ("text", "DOUBLE PRECISION"): _TEXT_TO_DOUBLE,
                                 ("text", "INTEGER"): "INTEGER USING NULLIF(\"{col}\", '')::integer",
+                                ("text", "BOOLEAN"): "BOOLEAN USING CASE WHEN \"{col}\" IN ('true', '1', 't', 'yes') THEN TRUE ELSE FALSE END",
                                 ("timestamp without time zone", "DOUBLE PRECISION"): 'DOUBLE PRECISION USING EXTRACT(EPOCH FROM "{col}" AT TIME ZONE \'UTC\')',
                                 ("timestamp with time zone", "DOUBLE PRECISION"): 'DOUBLE PRECISION USING EXTRACT(EPOCH FROM "{col}")',
                                 ("date", "DOUBLE PRECISION"): 'DOUBLE PRECISION USING EXTRACT(EPOCH FROM "{col}"::timestamp AT TIME ZONE \'UTC\')',

modules/connectors/connectorVoiceGoogle.py

@@ -19,6 +19,21 @@ from modules.shared.voiceCatalog import getDefaultVoice
 
 logger = logging.getLogger(__name__)
 
+_STT_LANGUAGE_MAP = {
+    "de-CH": "de-DE",
+    "de-AT": "de-DE",
+    "en-GB": "en-US",
+    "en-AU": "en-US",
+    "fr-CH": "fr-FR",
+    "it-CH": "it-IT",
+    "pt-BR": "pt-PT",
+}
+
+
+def _normalizeSttLanguage(language: str) -> str:
+    """Map regional language variants to codes supported by Google STT models."""
+    return _STT_LANGUAGE_MAP.get(language, language)
+
 
 def _buildPrimarySttRecognitionFields(
     *,
@@ -116,6 +131,7 @@ class ConnectorGoogleSpeech:
         Returns:
             Dict containing transcribed text, confidence, and metadata
         """
+        language = _normalizeSttLanguage(language)
         try:
             # Treat sampleRate=0 as unknown (invalid value from client)
             if sampleRate is not None and sampleRate <= 0:
@@ -480,6 +496,7 @@ class ConnectorGoogleSpeech:
             Dicts with keys: isFinal, transcript, confidence, stabilityScore, audioDurationSec;
             optionally endOfSingleUtterance, reconnectRequired
         """
+        language = _normalizeSttLanguage(language)
         STREAM_LIMIT_SEC = 290
         streamStartTs = time.time()
         totalAudioBytes = 0

modules/datamodels/datamodelSecurity.py

@@ -124,6 +124,43 @@ class Token(PowerOnModel):
         return data
 
 
+@i18nModel("Vertrauenswuerdiges Geraet")
+class TrustedDevice(PowerOnModel):
+    """A device trusted after successful MFA verification (skips MFA for configured duration)."""
+    id: str = Field(
+        default_factory=lambda: str(uuid.uuid4()),
+        description="Random token stored as httpOnly cookie value",
+        json_schema_extra={"label": "ID"},
+    )
+    userId: str = Field(
+        ...,
+        description="User this trusted device belongs to",
+        json_schema_extra={"label": "Benutzer-ID", "fk_target": {"db": "poweron_app", "table": "UserInDB", "labelField": "username"}},
+    )
+    trustedUntil: float = Field(
+        ...,
+        description="UTC timestamp until which the device is trusted",
+        json_schema_extra={"label": "Vertrauenswuerdig bis", "frontend_type": "timestamp"},
+    )
+    userAgent: Optional[str] = Field(
+        default=None,
+        description="Browser user agent at time of trust grant",
+        json_schema_extra={"label": "User-Agent"},
+    )
+    ipAddress: Optional[str] = Field(
+        default=None,
+        description="IP address at time of trust grant",
+        json_schema_extra={"label": "IP-Adresse"},
+    )
+    createdAt: float = Field(
+        default_factory=getUtcTimestamp,
+        description="When the device was trusted",
+        json_schema_extra={"label": "Erstellt am", "frontend_type": "timestamp"},
+    )
+
+    model_config = ConfigDict(use_enum_values=True)
+
+
 @i18nModel("Authentifizierungsereignis")
 class AuthEvent(PowerOnModel):
     """Authentication event for audit logging."""

modules/features/workspace/routeFeatureWorkspace.py

@@ -489,6 +489,158 @@ def _collectPriorFileIds(chatInterface, workflowId: str) -> List[str]:
     return result
 
 
+# Default context budget for prior files (metadata injection per file costs tokens
+# in EVERY agent round). Override per workspace instance via instanceConfig
+# key "maxPriorContextFiles". Files outside the budget remain fully accessible
+# to the agent via conversation history and readFile - nothing is lost.
+DEFAULT_MAX_PRIOR_CONTEXT_FILES = 10
+
+
+async def _selectPriorFilesForContext(
+    priorFileIds: List[str],
+    prompt: str,
+    aiObjects,
+    mandateId: str,
+    featureInstanceId: str,
+    maxFiles: int,
+) -> List[str]:
+    """Select which prior files get their metadata injected into the agent context.
+
+    Selection only happens when more candidates exist than the budget allows -
+    otherwise all files pass through untouched. Selection order:
+    1. Files whose indexed chunks are semantically relevant to the current prompt
+    2. Remaining budget filled with the most recent files (so files without an
+       index entry are never unfairly dropped)
+    """
+    if len(priorFileIds) <= maxFiles:
+        return priorFileIds
+
+    rankedFileIds: List[str] = []
+    try:
+        embeddingResponse = await aiObjects.callEmbedding([prompt])
+        embeddings = (embeddingResponse.metadata or {}).get("embeddings", [])
+        if embeddings:
+            knowledgeIf = getKnowledgeInterface()
+            results = knowledgeIf.semanticSearch(
+                queryVector=embeddings[0],
+                featureInstanceId=featureInstanceId,
+                mandateId=mandateId,
+                limit=maxFiles * 3,
+            )
+            priorSet = set(priorFileIds)
+            for chunk in results:
+                fid = chunk.get("fileId") if isinstance(chunk, dict) else getattr(chunk, "fileId", None)
+                if fid and fid in priorSet and fid not in rankedFileIds:
+                    rankedFileIds.append(fid)
+                    if len(rankedFileIds) >= maxFiles:
+                        break
+    except Exception as e:
+        logger.warning(f"Relevance ranking for prior files failed: {e}")
+
+    for fid in reversed(priorFileIds):
+        if len(rankedFileIds) >= maxFiles:
+            break
+        if fid not in rankedFileIds:
+            rankedFileIds.append(fid)
+
+    logger.info(
+        f"Prior-file selection: {len(priorFileIds)} candidates -> {len(rankedFileIds)} in context "
+        f"(budget={maxFiles}, relevance-ranked, recency fill)"
+    )
+    return rankedFileIds
+
+
+async def _ensureFilesIndexed(
+    fileIds: List[str],
+    user,
+    mandateId: str,
+    featureInstanceId: str,
+) -> int:
+    """Ensure all attached files have embeddings in the knowledge store.
+
+    Checks FileContentIndex for each file. Files not yet indexed are extracted
+    and embedded inline so that subsequent RAG queries can find their content.
+    Indexing is idempotent (content-hash check in requestIngestion), so each
+    file is only ever processed once - re-attaching an indexed file is a no-op.
+
+    Returns the number of files that were newly indexed.
+    """
+    if not fileIds:
+        return 0
+
+    knowledgeIf = getKnowledgeInterface(user)
+    unindexedIds = []
+    for fid in fileIds:
+        existing = knowledgeIf.getFileContentIndex(fid)
+        status = (existing.get("status") if isinstance(existing, dict) else getattr(existing, "status", "")) if existing else ""
+        if status not in ("indexed", "embedding"):
+            unindexedIds.append(fid)
+
+    if not unindexedIds:
+        return 0
+
+    logger.info(f"Ensure-embed: {len(unindexedIds)}/{len(fileIds)} files need indexing")
+
+    from modules.serviceCenter import getService
+    from modules.serviceCenter.context import ServiceCenterContext
+    indexCtx = ServiceCenterContext(
+        user=user, mandateId=mandateId, featureInstanceId=featureInstanceId,
+    )
+    try:
+        knowledgeService = getService("knowledge", indexCtx)
+    except Exception as e:
+        logger.warning(f"Ensure-embed: knowledge service unavailable: {e}")
+        return 0
+
+    chatInterface = interfaceDbChat.getInterface(user)
+    indexed = 0
+    for fid in unindexedIds:
+        try:
+            fileInfo = chatInterface.getFileInfo(fid) if chatInterface else None
+            if not fileInfo:
+                continue
+            fileName = fileInfo.get("fileName", "")
+            mimeType = fileInfo.get("mimeType", "")
+            rawBytes = chatInterface.getFileData(fid)
+            if not rawBytes:
+                continue
+
+            extractionService = getService("extraction", indexCtx)
+            extracted = extractionService.extractContentFromBytes(
+                rawBytes, fileName, mimeType, documentId=fid,
+            )
+            contentObjects = [
+                {
+                    "contentType": getattr(p, "contentType", "text"),
+                    "data": getattr(p, "data", "") or "",
+                    "contentObjectId": getattr(p, "contentObjectId", "") or str(uuid.uuid4()),
+                    "contextRef": getattr(p, "contextRef", {}) or {},
+                }
+                for p in (extracted.parts or [])
+                if getattr(p, "data", None)
+            ]
+            if not contentObjects:
+                continue
+
+            await knowledgeService.indexFile(
+                fileId=fid,
+                fileName=fileName,
+                mimeType=mimeType,
+                userId=user.id if user else "",
+                featureInstanceId=featureInstanceId,
+                mandateId=mandateId,
+                contentObjects=contentObjects,
+                structure=getattr(extracted, "structure", None),
+            )
+            indexed += 1
+        except Exception as e:
+            logger.debug(f"Ensure-embed: skipping file {fid}: {e}")
+
+    if indexed:
+        logger.info(f"Ensure-embed: indexed {indexed} file(s) before agent start")
+    return indexed
+
+
 async def _deriveWorkflowName(prompt: str, aiService) -> str:
     """Use AI to generate a concise workflow title from the user prompt."""
     from modules.datamodels.datamodelAi import AiCallRequest, AiCallOptions, OperationTypeEnum, PriorityEnum
@@ -740,20 +892,29 @@ async def _runWorkspaceAgent(
 
         priorFileIds = _collectPriorFileIds(chatInterface, workflowId)
         currentFileIdSet = set(fileIds or [])
-        mergedFileIds = list(fileIds or [])
-        for pf in priorFileIds:
-            if pf not in currentFileIdSet:
-                mergedFileIds.append(pf)
-        if len(mergedFileIds) > len(fileIds or []):
+        candidatePriorIds = [pf for pf in priorFileIds if pf not in currentFileIdSet]
+
+        # Embed-first rule: newly attached files are indexed into the knowledge
+        # store BEFORE the agent starts, so RAG retrieval works from round 1.
+        await _ensureFilesIndexed(fileIds or [], user, mandateId, instanceId)
+
+        _cfg = instanceConfig or {}
+
+        if candidatePriorIds:
+            maxPriorFiles = int(_cfg.get("maxPriorContextFiles", DEFAULT_MAX_PRIOR_CONTEXT_FILES))
+            candidatePriorIds = await _selectPriorFilesForContext(
+                candidatePriorIds, prompt, aiObjects, mandateId, instanceId, maxPriorFiles,
+            )
+
+        mergedFileIds = list(fileIds or []) + candidatePriorIds
+        if candidatePriorIds:
             logger.info(
-                f"Merged {len(mergedFileIds) - len(fileIds or [])} prior file(s) into agent context "
+                f"Merged {len(candidatePriorIds)} prior file(s) into agent context "
                 f"(total: {len(mergedFileIds)}) for workflow {workflowId}"
             )
 
         accumulatedText = ""
         messagePersisted = False
-
-        _cfg = instanceConfig or {}
         _toolSet = _cfg.get("toolSet", "core")
         _agentCfg = _cfg.get("agentConfig")
         from modules.serviceCenter.services.serviceAgent.datamodelAgent import AgentConfig

modules/interfaces/interfaceAiObjects.py

@@ -4,7 +4,9 @@ import logging
 import asyncio
 import uuid
 import base64
+import hashlib
 import json
+from collections import OrderedDict
 from typing import Dict, Any, List, Union, Tuple, Optional, Callable, AsyncGenerator
 from dataclasses import dataclass, field
 import time
@@ -542,7 +544,27 @@ class AiObjects:
         else:
             options.operationType = OperationTypeEnum.EMBEDDING
 
-        combinedText = " ".join(texts[:3])[:500]
+        # Serve known vectors from cache; only unknown texts go to the API.
+        resolvedVectors: Dict[int, List[float]] = {}
+        pendingTexts: List[str] = []
+        pendingPositions: List[int] = []
+        for i, t in enumerate(texts):
+            cachedVector = _embeddingCacheGet(t)
+            if cachedVector is not None:
+                resolvedVectors[i] = cachedVector
+            else:
+                pendingTexts.append(t)
+                pendingPositions.append(i)
+
+        if not pendingTexts:
+            logger.debug(f"Embedding cache hit for all {len(texts)} text(s)")
+            return AiCallResponse(
+                content="", modelName="embedding-cache", priceCHF=0.0,
+                processingTime=0.0, bytesSent=0, bytesReceived=0, errorCount=0,
+                metadata={"embeddings": [resolvedVectors[i] for i in range(len(texts))]},
+            )
+
+        combinedText = " ".join(pendingTexts[:3])[:500]
         availableModels = modelRegistry.getAvailableModels()
 
         allowedProviders = getattr(options, 'allowedProviders', None) if options else None
@@ -575,13 +597,13 @@ class AiObjects:
         for attempt, model in enumerate(failoverModelList):
             try:
                 logger.info(f"Embedding call with {model.name} (attempt {attempt + 1}/{len(failoverModelList)})")
-                inputBytes = sum(len(t.encode("utf-8")) for t in texts)
+                inputBytes = sum(len(t.encode("utf-8")) for t in pendingTexts)
                 startTime = time.time()
 
-                batches = _buildEmbeddingBatches(texts, model.contextLength)
+                batches = _buildEmbeddingBatches(pendingTexts, model.contextLength)
                 logger.info(
-                    f"Embedding: {len(texts)} texts -> {len(batches)} batch(es), "
-                    f"model contextLength={model.contextLength}"
+                    f"Embedding: {len(pendingTexts)} texts ({len(resolvedVectors)} cached) -> "
+                    f"{len(batches)} batch(es), model contextLength={model.contextLength}"
                 )
 
                 allEmbeddings: List[List[float]] = []
@@ -606,11 +628,17 @@ class AiObjects:
                 if totalPriceCHF == 0.0:
                     totalPriceCHF = model.calculatepriceCHF(processingTime, inputBytes, 0)
 
+                for j, position in enumerate(pendingPositions):
+                    if j < len(allEmbeddings):
+                        resolvedVectors[position] = allEmbeddings[j]
+                        _embeddingCachePut(pendingTexts[j], allEmbeddings[j])
+                mergedEmbeddings = [resolvedVectors.get(i, []) for i in range(len(texts))]
+
                 response = AiCallResponse(
                     content="", modelName=model.name, provider=model.connectorType,
                     priceCHF=totalPriceCHF, processingTime=processingTime,
                     bytesSent=inputBytes, bytesReceived=0, errorCount=0,
-                    metadata={"embeddings": allEmbeddings}
+                    metadata={"embeddings": mergedEmbeddings}
                 )
 
                 if self.billingCallback:
@@ -681,6 +709,28 @@ class AiObjects:
 _CHARS_PER_TOKEN = 4
 _SAFETY_MARGIN = 0.90
 
+# In-process cache for embedding vectors. Identical texts (e.g. the same user
+# prompt embedded once for prior-file selection and once for RAG context
+# building) hit the cache instead of paying for a second API call.
+_EMBEDDING_CACHE_MAX_ENTRIES = 256
+_embeddingCache: OrderedDict = OrderedDict()
+
+
+def _embeddingCacheGet(text: str) -> Optional[List[float]]:
+    key = hashlib.sha256(text.encode("utf-8")).hexdigest()
+    vector = _embeddingCache.get(key)
+    if vector is not None:
+        _embeddingCache.move_to_end(key)
+    return vector
+
+
+def _embeddingCachePut(text: str, vector: List[float]) -> None:
+    key = hashlib.sha256(text.encode("utf-8")).hexdigest()
+    _embeddingCache[key] = vector
+    _embeddingCache.move_to_end(key)
+    while len(_embeddingCache) > _EMBEDDING_CACHE_MAX_ENTRIES:
+        _embeddingCache.popitem(last=False)
+
 
 def _estimateTokens(text: str) -> int:
     """Rough token estimate: 1 token ~ 4 characters."""
@@ -691,9 +741,7 @@ def _buildEmbeddingBatches(texts: List[str], contextLength: int) -> List[List[st
     """Split a list of texts into batches whose total estimated token count
     stays within the model's contextLength (with safety margin).
 
-    Each individual text is assumed to already be within limits (enforced by
-    the chunking layer). If a single text exceeds the budget, it is placed
-    in its own batch as a last resort.
+    Texts that individually exceed the per-input limit are truncated to fit.
     """
     if not texts:
         return []
@@ -701,11 +749,21 @@ def _buildEmbeddingBatches(texts: List[str], contextLength: int) -> List[List[st
         return [texts]
 
     maxTokensPerBatch = int(contextLength * _SAFETY_MARGIN)
+    maxCharsPerInput = maxTokensPerBatch * _CHARS_PER_TOKEN
     batches: List[List[str]] = []
     currentBatch: List[str] = []
     currentTokens = 0
 
     for text in texts:
+        if len(text) > maxCharsPerInput:
+            # API hard limit per input. File content never hits this (the
+            # chunking layer splits at ~400 tokens); only oversized search
+            # queries can, where truncation is semantically acceptable.
+            logger.warning(
+                f"Embedding input truncated from {len(text)} to {maxCharsPerInput} chars "
+                f"(model input limit {contextLength} tokens)"
+            )
+            text = text[:maxCharsPerInput]
         textTokens = _estimateTokens(text)
         if currentBatch and (currentTokens + textTokens) > maxTokensPerBatch:
             batches.append(currentBatch)

modules/interfaces/interfaceDbApp.py

@@ -3112,8 +3112,12 @@ class AppObjects:
 
     # Token methods
 
-    def saveAccessToken(self, token: Token, replace_existing: bool = True) -> None:
-        """Save an access token for the current user (must NOT have connectionId)"""
+    def saveAccessToken(self, token: Token, replace_existing: bool = False) -> None:
+        """Save an access token for the current user (must NOT have connectionId).
+
+        Multi-session: replace_existing=False (default) keeps existing sessions alive.
+        Only set replace_existing=True for explicit single-session scenarios.
+        """
         try:
             # Validate that this is NOT a connection token
             if token.connectionId:

modules/routes/routeAdminSessions.py

@@ -0,0 +1,185 @@
+# Copyright (c) 2026 PowerOn AG
+# All rights reserved.
+"""
+Admin endpoints for session and trusted device management.
+
+Allows mandate-admins and platform-admins to view and revoke active sessions
+and trusted devices for users under their jurisdiction.
+"""
+
+from fastapi import APIRouter, HTTPException, status, Depends, Request, Query
+from typing import Dict, Any, List
+import logging
+
+from modules.auth import limiter, getCurrentUser
+from modules.datamodels.datamodelUam import User
+from modules.datamodels.datamodelSecurity import Token, TokenPurpose, TokenStatus, TrustedDevice
+from modules.interfaces.interfaceDbApp import getRootInterface
+from modules.shared.timeUtils import getUtcTimestamp
+from modules.shared.i18nRegistry import apiRouteContext
+
+routeApiMsg = apiRouteContext("routeAdminSessions")
+logger = logging.getLogger(__name__)
+
+router = APIRouter(
+    prefix="/api/admin/sessions",
+    tags=["Admin Sessions"],
+    responses={404: {"description": "Not found"}},
+)
+
+
+def _requireAdmin(currentUser: User) -> None:
+    """Ensure the caller is a platform admin or sysAdmin."""
+    if not (getattr(currentUser, "isPlatformAdmin", False) or getattr(currentUser, "isSysAdmin", False)):
+        raise HTTPException(
+            status_code=status.HTTP_403_FORBIDDEN,
+            detail=routeApiMsg("Only platform admins can manage sessions"),
+        )
+
+
+@router.get("")
+@limiter.limit("30/minute")
+def listSessions(
+    request: Request,
+    userId: str = Query(..., description="User ID whose sessions to list"),
+    currentUser: User = Depends(getCurrentUser),
+) -> List[Dict[str, Any]]:
+    """List active auth sessions for a user."""
+    _requireAdmin(currentUser)
+    rootInterface = getRootInterface()
+
+    tokens = rootInterface.db.getRecordset(
+        Token,
+        recordFilter={
+            "userId": userId,
+            "tokenPurpose": TokenPurpose.AUTH_SESSION.value,
+            "status": TokenStatus.ACTIVE.value,
+        },
+    )
+
+    now = getUtcTimestamp()
+    result = []
+    for t in tokens:
+        expiresAt = t.get("expiresAt", 0)
+        if expiresAt < now:
+            continue
+        result.append({
+            "sessionId": t.get("sessionId"),
+            "tokenId": t.get("id"),
+            "authority": t.get("authority"),
+            "createdAt": t.get("sysCreatedAt"),
+            "expiresAt": expiresAt,
+        })
+
+    return result
+
+
+@router.delete("/{sessionId}")
+@limiter.limit("30/minute")
+def revokeSession(
+    request: Request,
+    sessionId: str,
+    currentUser: User = Depends(getCurrentUser),
+) -> Dict[str, Any]:
+    """Revoke a single session by sessionId."""
+    _requireAdmin(currentUser)
+    rootInterface = getRootInterface()
+
+    tokens = rootInterface.db.getRecordset(
+        Token,
+        recordFilter={"sessionId": sessionId, "tokenPurpose": TokenPurpose.AUTH_SESSION.value},
+    )
+    count = 0
+    for t in tokens:
+        rootInterface.db.recordDelete(Token, t["id"])
+        count += 1
+
+    if count == 0:
+        raise HTTPException(status_code=404, detail=routeApiMsg("Session not found"))
+
+    logger.info(f"Admin {currentUser.username} revoked session {sessionId} ({count} token(s))")
+    return {"revoked": count, "sessionId": sessionId}
+
+
+@router.delete("")
+@limiter.limit("10/minute")
+def revokeAllSessions(
+    request: Request,
+    userId: str = Query(..., description="User ID whose sessions to revoke"),
+    currentUser: User = Depends(getCurrentUser),
+) -> Dict[str, Any]:
+    """Revoke ALL active sessions for a user (force logout everywhere)."""
+    _requireAdmin(currentUser)
+    rootInterface = getRootInterface()
+
+    tokens = rootInterface.db.getRecordset(
+        Token,
+        recordFilter={
+            "userId": userId,
+            "tokenPurpose": TokenPurpose.AUTH_SESSION.value,
+        },
+    )
+    count = 0
+    for t in tokens:
+        rootInterface.db.recordDelete(Token, t["id"])
+        count += 1
+
+    logger.info(f"Admin {currentUser.username} revoked all sessions for userId={userId} ({count} token(s))")
+    return {"revoked": count, "userId": userId}
+
+
+# --- Trusted Devices ---
+
+trustedDeviceRouter = APIRouter(
+    prefix="/api/admin/trusted-devices",
+    tags=["Admin Sessions"],
+    responses={404: {"description": "Not found"}},
+)
+
+
+@trustedDeviceRouter.get("")
+@limiter.limit("30/minute")
+def listTrustedDevices(
+    request: Request,
+    userId: str = Query(..., description="User ID whose trusted devices to list"),
+    currentUser: User = Depends(getCurrentUser),
+) -> List[Dict[str, Any]]:
+    """List trusted devices for a user."""
+    _requireAdmin(currentUser)
+    rootInterface = getRootInterface()
+
+    devices = rootInterface.db.getRecordset(
+        TrustedDevice, recordFilter={"userId": userId}
+    )
+
+    now = getUtcTimestamp()
+    result = []
+    for d in devices:
+        result.append({
+            "id": d.get("id", "")[:8] + "...",
+            "trustedUntil": d.get("trustedUntil"),
+            "isExpired": d.get("trustedUntil", 0) < now,
+            "userAgent": d.get("userAgent"),
+            "ipAddress": d.get("ipAddress"),
+            "createdAt": d.get("createdAt"),
+        })
+
+    return result
+
+
+@trustedDeviceRouter.delete("")
+@limiter.limit("10/minute")
+def revokeAllTrustedDevices(
+    request: Request,
+    userId: str = Query(..., description="User ID whose trusted devices to revoke"),
+    currentUser: User = Depends(getCurrentUser),
+) -> Dict[str, Any]:
+    """Revoke ALL trusted devices for a user (force MFA on next login)."""
+    _requireAdmin(currentUser)
+    rootInterface = getRootInterface()
+
+    from modules.auth.trustedDeviceService import revokeTrustedDevices
+    count = revokeTrustedDevices(userId, rootInterface.db)
+
+    logger.info(f"Admin {currentUser.username} revoked all trusted devices for userId={userId} ({count})")
+    return {"revoked": count, "userId": userId}

modules/routes/routeMfa.py

@@ -232,6 +232,13 @@ def mfaVerify(
 
     logger.info("MFA verify successful for user %s", username)
 
+    # Mark device as trusted so MFA is skipped on next login from this device
+    try:
+        from modules.auth.trustedDeviceService import createTrustedDevice
+        createTrustedDevice(userId, request, response, rootInterface.db)
+    except Exception as e:
+        logger.warning(f"Failed to create trusted device after MFA verify: {e}")
+
     try:
         from modules.dbHelpers.auditLogger import audit_logger
         audit_logger.logUserAccess(

modules/routes/routeSecurityLocal.py

@@ -256,6 +256,7 @@ def login(
 
         # --- MFA gate --------------------------------------------------------
         from modules.auth.mfaService import isMfaRequired as _isMfaRequired
+        from modules.auth.trustedDeviceService import isTrustedDevice as _isTrustedDevice
         from modules.routes.routeMfa import createMfaPendingToken
 
         userRecord = rootInterface._getUserForAuthentication(user.username)
@@ -273,7 +274,14 @@ def login(
         mfaRequired = _isMfaRequired(user, userMandates=userMandates, mandates=mandateObjs)
         hasMfaSetup = bool(userRecord and userRecord.get("mfaSecret") and getattr(user, "mfaEnabled", False))
 
+        # Trusted device: skip MFA if the device was previously verified
+        _deviceTrusted = False
         if mfaRequired or hasMfaSetup:
+            _deviceTrusted = _isTrustedDevice(request, str(user.id), rootInterface.db)
+            if _deviceTrusted:
+                logger.info(f"MFA skipped for user {user.username} (trusted device)")
+
+        if (mfaRequired or hasMfaSetup) and not _deviceTrusted:
             _sid = str(uuid.uuid4())
             pendingToken = createMfaPendingToken(
                 userId=str(user.id),
@@ -659,31 +667,46 @@ def refresh_token(
             logger.error(f"Failed to get user from database: {str(e)}")
             raise HTTPException(status_code=500, detail=routeApiMsg("Failed to validate user"))
         
+        # Preserve sessionId from the refresh token so the session stays grouped
+        sessionId = payload.get("sid") or str(uuid.uuid4())
+
         # Create new token data
-        # MULTI-TENANT: Token does NOT contain mandateId anymore
+        newJti = str(uuid.uuid4())
         token_data = {
             "sub": current_user.username,
             "userId": str(current_user.id),
-            "authenticationAuthority": current_user.authenticationAuthority
-            # NO mandateId in token
+            "authenticationAuthority": current_user.authenticationAuthority,
+            "jti": newJti,
+            "sid": sessionId,
         }
 
         # Create new access token + set cookie
-        access_token, _expires = createAccessToken(token_data)
+        access_token, accessExpires = createAccessToken(token_data)
         setAccessTokenCookie(response, access_token)
 
-        # Get expiration time
+        # Persist the new token in DB so _getUserBase() accepts it
+        authority = current_user.authenticationAuthority
+        if isinstance(authority, str):
+            authority = AuthAuthority(authority)
+        dbToken = Token(
+            id=newJti,
+            userId=str(current_user.id),
+            authority=authority,
+            tokenAccess=access_token,
+            tokenPurpose=TokenPurpose.AUTH_SESSION,
+            expiresAt=accessExpires.timestamp(),
+            sessionId=sessionId,
+        )
         try:
-            payload = jwt.decode(access_token, SECRET_KEY, algorithms=[ALGORITHM])
-            expires_at = datetime.fromtimestamp(payload.get("exp"))
+            userInterface = getInterface(current_user)
+            userInterface.saveAccessToken(dbToken)
         except Exception as e:
-            logger.error(f"Failed to decode new access token: {str(e)}")
-            raise HTTPException(status_code=500, detail=routeApiMsg("Failed to create new token"))
+            logger.warning(f"Failed to persist refreshed token in DB: {e}")
 
         return {
             "type": "token_refresh_success",
             "message": "Token refreshed successfully",
-            "expires_at": expires_at.isoformat()
+            "expires_at": accessExpires.isoformat()
         }
         
     except HTTPException as e:

modules/serviceCenter/services/serviceKnowledge/mainServiceKnowledge.py

@@ -638,7 +638,8 @@ class KnowledgeService:
         Returns:
             Formatted context string for injection into the agent's system prompt.
         """
-        queryVector = await self._embedSingle(currentPrompt)
+        queryText = _extractUserQuery(currentPrompt)
+        queryVector = await self._embedSingle(queryText) if queryText else []
         logger.debug(
             "buildAgentContext.start userId=%s featureInstanceId=%s mandateId=%s isSysAdmin=%s prompt=%r",
             userId, featureInstanceId, mandateId, isSysAdmin, (currentPrompt or "")[:120],
@@ -960,6 +961,29 @@ class KnowledgeService:
 # Internal helpers
 # =============================================================================
 
+# Markers added by the prompt enrichment layers (_enrichPromptWithFiles in
+# mainServiceAgent, data source sections in routeFeatureWorkspace). Used to
+# isolate the user's own words for the semantic search query.
+_USER_REQUEST_MARKER = "\n\nUser request: "
+_PROMPT_SECTION_MARKERS = ("\n\n[Active Data Sources]\n", "\n\n[Attached Feature Data Sources]\n")
+
+
+def _extractUserQuery(prompt: str) -> str:
+    """Isolate the user's actual question from an enriched agent prompt.
+
+    Enriched prompts wrap the user request in file metadata headers and
+    data-source sections. Only the user's own words form a meaningful semantic
+    search query - embedding the metadata would dilute the vector and can
+    exceed the embedding model's input limit.
+    """
+    if not prompt:
+        return ""
+    text = prompt.rsplit(_USER_REQUEST_MARKER, 1)[-1]
+    for marker in _PROMPT_SECTION_MARKERS:
+        text = text.split(marker, 1)[0]
+    return text.strip()
+
+
 def _estimateTokens(text: str) -> int:
     """Estimate token count using character-based heuristic (1 token ~ 4 chars)."""
     return max(1, len(text) // CHARS_PER_TOKEN)