File size: 45,568 Bytes
b02630d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6ceb8b5
3021956
 
4351cdd
3021956
 
 
6ceb8b5
 
 
 
 
 
 
 
 
 
d76cab0
 
6ceb8b5
3021956
 
b02630d
d76cab0
b02630d
d76cab0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b02630d
 
 
 
 
d76cab0
 
 
 
 
 
 
 
b02630d
d76cab0
 
 
 
 
 
 
 
b02630d
 
4351cdd
 
 
 
d76cab0
4351cdd
 
 
 
 
 
 
 
 
 
d76cab0
b02630d
 
 
 
 
4351cdd
b02630d
4351cdd
 
 
b02630d
 
 
 
4351cdd
b02630d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
93e0b15
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b02630d
 
93e0b15
 
 
b02630d
93e0b15
 
 
 
 
 
 
 
b02630d
93e0b15
b02630d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4351cdd
 
 
2fccaae
 
 
b02630d
c17f43c
 
b02630d
2fccaae
b02630d
 
 
 
 
 
 
 
2fccaae
b02630d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c17f43c
 
 
 
 
 
 
 
 
 
 
 
 
 
b02630d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c17f43c
 
 
 
 
 
 
 
 
 
 
 
 
 
b02630d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c17f43c
4351cdd
 
 
 
 
c17f43c
 
 
 
 
b02630d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c17f43c
4351cdd
 
 
 
 
 
 
c17f43c
 
 
 
 
 
 
b02630d
2fccaae
 
 
 
b02630d
2fccaae
b02630d
 
 
 
 
 
 
 
 
 
 
 
714f42f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8e5db5f
714f42f
b47dcd2
 
 
 
 
 
714f42f
 
 
 
 
34a913a
714f42f
 
 
 
 
8e5db5f
34a913a
714f42f
 
 
 
34a913a
714f42f
 
 
 
34a913a
714f42f
34a913a
 
714f42f
34a913a
714f42f
 
 
 
 
 
 
 
34a913a
 
 
 
 
 
 
b47dcd2
714f42f
34a913a
8e5db5f
b47dcd2
8e5db5f
34a913a
 
 
 
 
 
 
 
b47dcd2
8e5db5f
 
 
 
34a913a
8e5db5f
34a913a
b47dcd2
8e5db5f
 
b47dcd2
8e5db5f
34a913a
b47dcd2
34a913a
b47dcd2
8e5db5f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
714f42f
8e5db5f
b47dcd2
 
 
 
8e5db5f
b47dcd2
 
714f42f
b47dcd2
 
714f42f
b47dcd2
714f42f
b47dcd2
 
 
 
 
714f42f
b47dcd2
 
 
714f42f
b47dcd2
 
714f42f
b47dcd2
 
 
 
 
 
8e5db5f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b47dcd2
8e5db5f
b47dcd2
8e5db5f
b47dcd2
 
 
 
 
8e5db5f
 
 
 
 
 
 
714f42f
b47dcd2
 
 
 
 
 
 
8e5db5f
 
b47dcd2
 
 
 
 
 
 
 
 
8e5db5f
b47dcd2
8e5db5f
b47dcd2
714f42f
8e5db5f
 
 
 
 
714f42f
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
# ===================== api.py ==========================
from typing import List, Dict
from pathlib import Path
from fastapi import FastAPI, UploadFile, File, Form
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import StreamingResponse
from pydantic import BaseModel

from config.config import Config
from config.system_prompt import PPLX_SYSTEM_PROMPT

# Core routing / graph
from rag.router import RouterAgent
from rag.graph_deep import (
    DeepResearchGraph,
    WebSearchGraph,
    RAGOnlyGraph,
    AgenticRAGGraph,
    AnalysisGraph,
    SummarizeGraph
)

# Tools
from tools.memory_tool import MemoryTool
from tools.name_tool import NameTool
from tools.search_tool import SearchTool
from tools.browse_tool import BrowseTool
from tools.reranker_tool import Reranker
from tools.followup_tool import FollowUpGenerator
from tools.image_tavily import TavilyImageSearch
from tools.knowledge_panel import KnowledgePanel
from tools.summarizer_tool import SummarizerTool

# RAG pipeline
from document_processing.processor import DocumentProcessor
from vectorstore.store import VectorStore

# File Manager for per-workspace RAG
from files.file_manager import FileManager


# =======================================================
# FastAPI App
# =======================================================
app = FastAPI(title="Perplexity Clone API", version="8.0 - Production LangGraph")

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_headers=["*"],
    allow_methods=["*"],
    allow_credentials=True,
)

import os

# =======================================================
# Health Check Endpoint
# =======================================================
@app.get("/health")
async def health_check():
    """Health check endpoint - shows API key status"""
    groq_key = os.getenv("GROQ_API_KEY", "")
    tavily_key = os.getenv("TAVILY_API_KEY", "")
    
    return {
        "status": "healthy",
        "service": "perplexity-clone-api",
        "api_keys": {
            "groq": "configured" if groq_key else "MISSING",
            "tavily": "configured" if tavily_key else "MISSING"
        },
        "model": Config.LLM_MODEL
    }


# =======================================================
# Global Components - with startup logging
# =======================================================
print("\n" + "="*50)
print("πŸš€ PERPLEXITY CLONE API STARTING UP")
print("="*50)

# Check API keys at startup
_groq_key = os.getenv("GROQ_API_KEY", "")
_tavily_key = os.getenv("TAVILY_API_KEY", "")
print(f"πŸ“‹ GROQ_API_KEY: {'βœ… Configured' if _groq_key else '❌ MISSING'}")
print(f"πŸ“‹ TAVILY_API_KEY: {'βœ… Configured' if _tavily_key else '❌ MISSING'}")

try:
    llm = Config.get_llm()
    print("βœ… LLM initialized successfully")
except Exception as e:
    print(f"❌ LLM initialization failed: {e}")
    llm = None

router = RouterAgent()

memory = MemoryTool()
name_tool = NameTool()
followup = FollowUpGenerator()

try:
    search_tool = SearchTool()
    print("βœ… SearchTool initialized")
except Exception as e:
    print(f"⚠️ SearchTool init error: {e}")
    search_tool = None

browse_tool = BrowseTool()

try:
    image_search = TavilyImageSearch()
    print("βœ… ImageSearch initialized")
except Exception as e:
    print(f"⚠️ ImageSearch init error: {e}")
    image_search = None

summarizer = SummarizerTool()

# Load all components
reranker = Reranker()
knowledge_panel = KnowledgePanel()

print("\nπŸ“š Loading demo documents...")
# RAG demo vectorstore
processor = DocumentProcessor(
    chunk_size=Config.CHUNK_SIZE,
    chunk_overlap=Config.CHUNK_OVERLAP,
)
demo_docs = processor.load_url("https://lilianweng.github.io/posts/2023-06-23-agent/")
demo_splits = processor.split(demo_docs)

vector = VectorStore()
vector.create(demo_splits)
print("βœ… Vector store initialized")

# File manager for per-workspace document RAG
file_manager = FileManager(base_dir="workspace_data")

# =======================================================
# Initialize All LangGraph Pipelines
# =======================================================
deep_graph = DeepResearchGraph(vector)
rag_graph = RAGOnlyGraph(file_manager)
agentic_graph = AgenticRAGGraph(file_manager, vector, image_search)
web_graph = WebSearchGraph()
analysis_graph = AnalysisGraph()
summarize_graph = SummarizeGraph()

print("βœ… All LangGraph pipelines initialized!")


# =======================================================
# Models
# =======================================================
class ChatRequest(BaseModel):
    message: str
    workspace_id: str = "default"


class ChatResponse(BaseModel):
    answer: str
    sources: List[Dict[str, str]] = []
    links: List[Dict[str, str]] = []
    images: List[Dict[str, str]] = []
    followups: List[str] = []
    default_tab: str = "answer"  # "answer" | "links" | "images"
    workspace_id: str


# =======================================================
# Utils
# =======================================================
def build_context(ws: str, new_msg: str):
    """Inject full workspace chat history + system prompt."""
    messages = [{"role": "system", "content": PPLX_SYSTEM_PROMPT}]
    for msg in memory.get_long_chat(ws):
        messages.append({"role": msg["role"], "content": msg["content"]})
    messages.append({"role": "user", "content": new_msg})
    return messages


def guess_default_tab(query: str, mode: str) -> str:
    """Decide which UI tab should be first (Answer / Links / Images)."""
    q = query.lower()
    image_words = [
        "image", "images", "photo", "photos", "picture", "pictures",
        "wallpaper", "logo", "flag", "screenshot", "pic"
    ]
    if any(w in q for w in image_words):
        return "images"
    if mode == "web":
        return "links"
    return "answer"


def tavily_images_safe(query: str) -> List[Dict[str, str]]:
    """Safe wrapper around Tavily image search."""
    try:
        return image_search.search(query, count=6)
    except Exception as e:
        print("Tavily image search error:", e)
        return []


def convert_links(results: List[Dict]) -> List[Dict[str, str]]:
    """Convert Tavily web search results to link objects."""
    links = []
    for r in results:
        url = r.get("url")
        if not url:
            continue
        links.append(
            {
                "title": r.get("title", "Result"),
                "url": url,
                "snippet": (r.get("content") or "")[:200],
            }
        )
    return links


# =======================================================
# Chat Endpoint
# =======================================================
@app.post("/api/chat", response_model=ChatResponse)
def chat(req: ChatRequest):

    q = req.message.strip()
    ws = req.workspace_id

    memory.add(ws, "user", q)

    # -------- Name memory special cases --------
    extracted = name_tool.extract_name(q)
    if extracted:
        memory.set_name(ws, extracted)
        reply = f"Nice to meet you, {extracted}! I’ll remember your name."
        memory.add(ws, "assistant", reply)
        return ChatResponse(answer=reply, workspace_id=ws)

    if q.lower() in ["tell me my name", "what is my name"]:
        nm = memory.get_name(ws)
        ans = f"Your name is {nm} 😊" if nm else "You haven’t told me your name yet."
        memory.add(ws, "assistant", ans)
        return ChatResponse(answer=ans, workspace_id=ws)

    # -------- Routing --------
    mode = router.route(q)
    default_tab = guess_default_tab(q, mode)

    answer = ""
    links: List[Dict[str, str]] = []
    sources: List[Dict[str, str]] = []
    follow: List[str] = []

    # -------- LLM Mode (chat / creative / small talk) --------
    if mode == "llm":
        msgs = build_context(ws, q)
        answer = llm.invoke(msgs).content
        follow = followup.generate(answer, q)

        # Optional small set of links
        try:
            res = search_tool.search(q, num_results=3)
            links = convert_links(res)
        except Exception as e:
            print("search error (llm mode):", e)

    # -------- Image Mode (image search queries) --------
    elif mode == "image":
        # For image queries, provide brief context + focus on images tab
        try:
            res = search_tool.search(q, num_results=3)
            ctx = res[0].get("snippet", "") if res else ""
            answer = f"Here are images related to '{q}'."
            if ctx:
                answer += f"\n\n{ctx}"
            links = convert_links(res)
        except Exception as e:
            print("search error (image mode):", e)
            answer = f"Showing images for: {q}"
        
        follow = []

    # -------- AGENTIC RAG Mode (files + web + images + knowledge) --------
    elif mode == "rag":
        # PLANNER AGENT: Decide which agents to activate
        q_lower = q.lower()
        
        use_file_rag = any(
            w in q_lower for w in [
                "summarize", "according to", "in this pdf", "in the document",
                "based on the file", "read my", "extract from", "uploaded",
                "this file", "the file", "my file", "from file"
            ]
        ) or len(q.split()) > 2  # multi-word questions likely need file RAG
        
        use_web = any(
            w in q_lower for w in [
                "today", "latest", "current", "news", "stock", "price",
                "real-time", "weather", "who is", "what is", "where is",
                "when", "how much", "compare"
            ]
        )
        
        use_images = any(
            w in q_lower for w in [
                "image", "images", "logo", "flag", "photos", "look like", 
                "picture", "show me", "wallpaper", "screenshot"
            ]
        )
        
        # FILE AGENT: Retrieve from workspace uploaded docs
        ws_obj = file_manager.get_workspace(ws)
        file_chunks = []
        if use_file_rag and ws_obj.initialized:
            file_chunks = ws_obj.retrieve(q, k=6)
        
        # REFERENCE AGENT: Retrieve from base vector store (demo docs)
        base_chunks = vector.retrieve(q, k=4)
        base_chunks = reranker.rerank(q, base_chunks, top_k=3)
        
        # WEB AGENT: Fetch live web content
        web_pages = []
        web_results = []
        if use_web:
            try:
                web_results = search_tool.search(q, num_results=4)
                for r in web_results:
                    url = r.get("url")
                    if not url:
                        continue
                    text = browse_tool.fetch_clean(url)
                    if text:
                        web_pages.append({
                            "title": r.get("title", ""),
                            "url": url,
                            "content": text[:1500]  # Speed optimization
                        })
            except Exception as e:
                print(f"Web agent error: {e}")
        
        # IMAGE AGENT: Fetch relevant images
        images_result = tavily_images_safe(q) if use_images else []
        
        # BUILD COMBINED CONTEXT
        contexts = []
        
        if file_chunks:
            file_ctx = "\n\n".join(d.page_content for d in file_chunks)
            contexts.append(f"πŸ“„ FILE CONTEXT (from uploaded documents):\n{file_ctx}")
        
        if base_chunks:
            ref_ctx = "\n\n".join(d.page_content for d in base_chunks)
            contexts.append(f"πŸ“š REFERENCE CONTEXT:\n{ref_ctx}")
        
        if web_pages:
            web_ctx = "\n\n".join(f"[{p['title']}]: {p['content']}" for p in web_pages)
            contexts.append(f"🌐 WEB CONTEXT (live web data):\n{web_ctx}")
        
        full_context = "\n\n-----\n\n".join(contexts) if contexts else "No context available."
        
        # SYNTHESIZER AGENT: Generate final answer
        synth_prompt = f"""You are an AGENTIC RAG synthesis model like Perplexity AI.
Combine information from FILE CONTEXT, REFERENCE CONTEXT and WEB CONTEXT.

RULES:
1. PRIORITIZE info from FILE CONTEXT (user's uploaded documents) when available.
2. Use WEB CONTEXT to add current/live information.
3. Use REFERENCE CONTEXT for background knowledge.
4. Cite sources using [1], [2], etc. when referencing specific info.
5. If answering from a file, say "According to your uploaded document..."
6. Do NOT hallucinate - only use info from the provided contexts.
7. Be concise but comprehensive.

AVAILABLE CONTEXT:
{full_context}

USER QUESTION: {q}

FINAL ANSWER:"""
        
        msgs = build_context(ws, synth_prompt)
        answer = llm.invoke(msgs).content
        follow = followup.generate(answer, q)
        
        # BUILD SOURCES
        sources = []
        if file_chunks:
            for d in file_chunks:
                sources.append({
                    "title": d.metadata.get("source", "πŸ“„ Uploaded File"),
                    "url": d.metadata.get("file_path", "")
                })
        
        if web_pages:
            for p in web_pages:
                sources.append({"title": p["title"], "url": p["url"]})
        
        # BUILD LINKS
        links = convert_links(web_results)
        
        # Set images from image agent
        if images_result:
            # Will be set at the end with tavily_images_safe
            pass

    # -------- Web Mode (real-time / entities / news) --------
    elif mode == "web":
        # Use WebSearchGraph for proper web search
        try:
            state = web_graph.run(q)
            answer = state.get("answer", "No answer generated.")
            sources = state.get("sources", [])
            links = state.get("links", [])
            follow = state.get("followups", [])
        except Exception as e:
            print(f"Web search error in chat: {e}")
            # Fallback to direct search
            res = search_tool.search(q, num_results=5)

            pages = []
            for r in res:
                url = r.get("url")
                if not url:
                    continue
                text = browse_tool.fetch_clean(url)
                if not text:
                    continue
                pages.append(
                    {
                        "title": r.get("title", "Webpage"),
                        "url": url,
                        "content": text[:2000],
                    }
                )

            ctx = "\n\n".join(p["content"] for p in pages)
            prompt = (
                "Use ONLY the following web content to answer. "
                "Cite sources using [1], [2], etc.\n\n"
                f"{ctx}\n\nQuestion: {q}"
            )

            msgs = build_context(ws, prompt)
            answer = llm.invoke(msgs).content
            follow = followup.generate(answer, q)

            links = [
                {
                    "title": p["title"],
                    "url": p["url"],
                    "snippet": p["content"][:200],
                }
                for p in pages
            ]

            sources = [{"title": p["title"], "url": p["url"]} for p in pages]

    # -------- Fallback β†’ LLM --------
    else:
        msgs = build_context(ws, q)
        answer = llm.invoke(msgs).content
        follow = followup.generate(answer, q)

    # -------- Images (for Images tab) --------
    images = tavily_images_safe(q)
    
    # Debug logging
    print(f"\n=== API Response Debug ===")
    print(f"Mode: {mode}")
    print(f"Links count: {len(links)}")
    print(f"Images count: {len(images)}")
    print(f"Sources count: {len(sources)}")
    if links:
        print(f"First link: {links[0]}")
    if images:
        print(f"First image: {images[0]}")
    print(f"========================\n")

    memory.add(ws, "assistant", answer)

    return ChatResponse(
        answer=answer,
        sources=sources,
        links=links,
        images=images,
        followups=follow,
        default_tab=default_tab,
        workspace_id=ws,
    )


# =======================================================
# Streaming Endpoint
# =======================================================
@app.post("/api/chat/stream")
def chat_stream(req: ChatRequest):

    q = req.message
    ws = req.workspace_id
    memory.add(ws, "user", q)

    msgs = build_context(ws, q)

    def generate():
        full = ""
        for chunk in llm.stream(msgs):
            tok = getattr(chunk, "content", "")
            if tok:
                full += tok
                yield tok
        memory.add(ws, "assistant", full)

    return StreamingResponse(generate(), media_type="text/plain")


# =======================================================
# Deep Research Endpoint
# =======================================================
@app.post("/api/deep_research", response_model=ChatResponse)
def deep_research(req: ChatRequest):

    q = req.message
    ws = req.workspace_id

    memory.add(ws, "user", q)

    try:
        state = deep_graph.run(q)
        answer = state.get("final_answer", "No answer generated.")
        sources = state.get("sources", [])
        # Get links from web_pages
        web_pages = state.get("web_pages", [])
        links = [{"title": p.get("title", ""), "url": p.get("url", ""), "snippet": p.get("content", "")[:200]} for p in web_pages]
    except Exception as e:
        print(f"Deep research error: {e}")
        answer = f"Deep research encountered an error. Please try again."
        sources = []
        links = []

    memory.add(ws, "assistant", answer)
    images = tavily_images_safe(q)
    follow = followup.generate(answer, q)

    return ChatResponse(
        answer=answer,
        sources=sources,
        links=links,
        images=images,
        followups=follow,
        default_tab="answer",
        workspace_id=ws,
    )


# =======================================================
# Knowledge Panel Endpoint
# =======================================================
@app.get("/api/knowledge_panel")
def get_knowledge_panel(q: str):
    """
    Returns Wikipedia-style infobox + AI-generated facts.
    Used by UI to render a sidebar knowledge card.
    """
    try:
        panel = knowledge_panel.build_panel(q)
        return panel
    except Exception as e:
        print("Knowledge panel error:", e)
        return {"wiki": {}, "facts": []}


# =======================================================
# FILE UPLOAD (PDF / TXT / PPTX) - Perplexity Spaces Feature
# =======================================================
@app.post("/api/upload_docs")
async def upload_docs(
    workspace_id: str = Form("default"),
    files: List[UploadFile] = File(...)
):
    """
    Upload one or more documents and index them for this workspace.
    Supports PDF, TXT, MD, PPT, PPTX files.
    """
    ws = file_manager.get_workspace(workspace_id)
    saved_paths = []

    for f in files:
        ext = Path(f.filename).suffix.lower()
        if ext not in [".pdf", ".txt", ".md", ".ppt", ".pptx"]:
            continue  # skip unsupported types

        dest = Path(ws.base_dir) / f.filename
        with open(dest, "wb") as out:
            content = await f.read()
            out.write(content)
        saved_paths.append(dest)

    if saved_paths:
        ws.add_files(saved_paths)
        print(f"βœ… Indexed {len(saved_paths)} files for workspace '{workspace_id}'")

    return {
        "workspace_id": workspace_id,
        "files": ws.files,
        "count": len(ws.files),
        "message": f"Successfully indexed {len(saved_paths)} files"
    }


@app.get("/api/workspace_files/{workspace_id}")
def get_workspace_files(workspace_id: str):
    """Get list of files uploaded to a workspace."""
    ws = file_manager.get_workspace(workspace_id)
    return {
        "workspace_id": workspace_id,
        "files": ws.files,
        "initialized": ws.initialized
    }


@app.delete("/api/workspace/{workspace_id}")
def clear_workspace(workspace_id: str):
    """Clear all files from a workspace."""
    file_manager.clear_workspace(workspace_id)
    return {"message": f"Workspace '{workspace_id}' cleared"}


# =======================================================
# MODE-SPECIFIC ENDPOINTS
# =======================================================

class ModeRequest(BaseModel):
    message: str
    workspace_id: str = "default"
    mode: str = "auto"


@app.post("/api/focus", response_model=ChatResponse)
def focus_mode(req: ModeRequest):
    """Focus mode - concise, direct answers without web search."""
    q = req.message.strip()
    ws = req.workspace_id
    
    memory.add(ws, "user", q)
    
    prompt = f"""You are in FOCUS mode. Provide a concise, direct answer.
- No unnecessary elaboration
- Get straight to the point
- Use bullet points if helpful
- Be accurate and helpful

Question: {q}

Answer:"""
    
    msgs = build_context(ws, prompt)
    answer = llm.invoke(msgs).content
    follow = followup.generate(answer, q)
    
    memory.add(ws, "assistant", answer)
    
    return ChatResponse(
        answer=answer,
        sources=[],
        links=[],
        images=[],
        followups=follow,
        default_tab="answer",
        workspace_id=ws
    )


@app.post("/api/writing", response_model=ChatResponse)
def writing_mode(req: ModeRequest):
    """Writing mode - creative writing, essays, content generation."""
    q = req.message.strip()
    ws = req.workspace_id
    
    memory.add(ws, "user", q)
    
    prompt = f"""You are in WRITING mode - a creative writing assistant.
Help with:
- Essays, articles, blog posts
- Creative writing, stories
- Professional emails and documents
- Content improvement and editing
- Grammar and style suggestions

Be creative, engaging, and helpful. Format your response well.

Request: {q}

Response:"""
    
    msgs = build_context(ws, prompt)
    answer = llm.invoke(msgs).content
    follow = followup.generate(answer, q)
    
    memory.add(ws, "assistant", answer)
    
    return ChatResponse(
        answer=answer,
        sources=[],
        links=[],
        images=[],
        followups=follow,
        default_tab="answer",
        workspace_id=ws
    )


@app.post("/api/math", response_model=ChatResponse)
def math_mode(req: ModeRequest):
    """Math mode - mathematical calculations and explanations."""
    q = req.message.strip()
    ws = req.workspace_id
    
    memory.add(ws, "user", q)
    
    prompt = f"""You are in MATH mode - a mathematical assistant.
- Solve mathematical problems step by step
- Show all work and calculations
- Explain the reasoning
- Use proper mathematical notation
- Handle algebra, calculus, statistics, geometry, etc.

Problem: {q}

Solution:"""
    
    msgs = build_context(ws, prompt)
    answer = llm.invoke(msgs).content
    follow = followup.generate(answer, q)
    
    memory.add(ws, "assistant", answer)
    
    return ChatResponse(
        answer=answer,
        sources=[],
        links=[],
        images=[],
        followups=follow,
        default_tab="answer",
        workspace_id=ws
    )


@app.post("/api/code", response_model=ChatResponse)
def code_mode(req: ModeRequest):
    """Code mode - programming help and code generation."""
    q = req.message.strip()
    ws = req.workspace_id
    
    memory.add(ws, "user", q)
    
    prompt = f"""You are in CODE mode - an expert programming assistant.
- Write clean, efficient, well-commented code
- Explain the code logic
- Follow best practices
- Handle any programming language
- Debug and fix code issues
- Suggest improvements

Request: {q}

Response:"""
    
    msgs = build_context(ws, prompt)
    answer = llm.invoke(msgs).content
    follow = followup.generate(answer, q)
    
    memory.add(ws, "assistant", answer)
    
    return ChatResponse(
        answer=answer,
        sources=[],
        links=[],
        images=[],
        followups=follow,
        default_tab="answer",
        workspace_id=ws
    )


@app.post("/api/analyze", response_model=ChatResponse)
def analyze_mode(req: ModeRequest):
    """
    Analysis mode - deep analysis with web research.
    Production-level LangGraph implementation.
    """
    q = req.message.strip()
    ws = req.workspace_id
    
    memory.add(ws, "user", q)
    
    try:
        # Run the AnalysisGraph pipeline
        state = analysis_graph.run(q)
        
        answer = state.get("answer", "No analysis generated.")
        sources = state.get("sources", [])
        links = state.get("links", [])
        follow = state.get("followups", [])
    except Exception as e:
        print(f"Analysis error: {e}")
        answer = f"Analysis encountered an error: {str(e)[:100]}"
        sources = []
        links = []
        follow = []
    
    # Get related images
    images = tavily_images_safe(q)
    
    memory.add(ws, "assistant", answer)
    
    return ChatResponse(
        answer=answer,
        sources=sources,
        links=links,
        images=images,
        followups=follow,
        default_tab="answer",
        workspace_id=ws
    )


@app.post("/api/summarize", response_model=ChatResponse)
def summarize_mode(req: ModeRequest):
    """
    Summarize mode - summarize uploaded documents OR web content.
    Prioritizes uploaded files, then falls back to web search.
    """
    q = req.message.strip()
    ws = req.workspace_id
    
    memory.add(ws, "user", q)
    
    # STEP 1: Check for uploaded files first
    ws_obj = file_manager.get_workspace(ws)
    
    if ws_obj.initialized and ws_obj.files:
        # Summarize from uploaded files
        print(f"πŸ“ SUMMARIZE MODE: Using uploaded files")
        try:
            # Retrieve relevant chunks from files
            chunks = ws_obj.retrieve(q, k=10)
            if chunks:
                # Combine chunk content for summarization
                content = "\n\n".join([c.page_content for c in chunks])
                
                # Generate summary
                summary = summarizer.summarize(content, max_words=400)
                
                # Build sources from files
                seen_files = set()
                sources = []
                for c in chunks:
                    fname = c.metadata.get("source", "Document")
                    if fname not in seen_files:
                        sources.append({"title": f"πŸ“„ {fname}", "url": ""})
                        seen_files.add(fname)
                
                follow = followup.generate(summary, q)
                
                memory.add(ws, "assistant", summary)
                
                return ChatResponse(
                    answer=summary,
                    sources=sources,
                    links=[],
                    images=[],
                    followups=follow,
                    default_tab="answer",
                    workspace_id=ws
                )
        except Exception as e:
            print(f"  ❌ File summarize error: {e}")
    
    # STEP 2: Check if it's a URL
    if q.startswith("http"):
        print(f"πŸ“ SUMMARIZE MODE: URL detected")
        try:
            content = browse_tool.fetch_clean(q)
            if content:
                summary = summarizer.summarize(content, max_words=400)
                sources = [{"title": "Source URL", "url": q}]
                links = [{"title": "Source", "url": q, "snippet": content[:200]}]
                follow = followup.generate(summary, q)
                
                memory.add(ws, "assistant", summary)
                
                return ChatResponse(
                    answer=summary,
                    sources=sources,
                    links=links,
                    images=[],
                    followups=follow,
                    default_tab="answer",
                    workspace_id=ws
                )
        except Exception as e:
            print(f"  ❌ URL fetch error: {e}")
    
    # STEP 3: Fall back to web search and summarize
    print(f"πŸ“ SUMMARIZE MODE: Web search fallback")
    try:
        results = search_tool.search(q, num_results=3)
        content_parts = []
        links = []
        
        for r in results:
            url = r.get("url", "")
            title = r.get("title", "")
            text = browse_tool.fetch_clean(url)
            if text:
                content_parts.append(text[:1500])
                links.append({"title": title, "url": url, "snippet": text[:150]})
        
        if content_parts:
            combined = "\n\n".join(content_parts)
            summary = summarizer.summarize(combined, max_words=400)
        else:
            summary = "Could not find content to summarize."
        
        sources = [{"title": l["title"], "url": l["url"]} for l in links]
        follow = followup.generate(summary, q)
        
        memory.add(ws, "assistant", summary)
        
        return ChatResponse(
            answer=summary,
            sources=sources,
            links=links,
            images=[],
            followups=follow,
            default_tab="answer",
            workspace_id=ws
        )
    except Exception as e:
        print(f"  ❌ Summarize error: {e}")
        return ChatResponse(
            answer=f"Error generating summary: {str(e)}",
            sources=[],
            links=[],
            images=[],
            followups=[],
            default_tab="answer",
            workspace_id=ws
        )


# =======================================================
# PRODUCTION-LEVEL MODE ENDPOINTS
# =======================================================

@app.post("/api/web", response_model=ChatResponse)
def web_search_mode(req: ModeRequest):
    """
    Web Search Mode - Real-time web search with source citations.
    Production-level LangGraph implementation.
    """
    q = req.message.strip()
    ws = req.workspace_id
    
    memory.add(ws, "user", q)
    
    try:
        # Run the WebSearchGraph pipeline
        state = web_graph.run(q)
        
        answer = state.get("answer", "No answer generated.")
        sources = state.get("sources", [])
        links = state.get("links", [])
        follow = state.get("followups", [])
    except Exception as e:
        print(f"Web search error: {e}")
        answer = f"Web search encountered an error: {str(e)[:100]}"
        sources = []
        links = []
        follow = []
    
    # Get images separately
    images = tavily_images_safe(q)
    
    memory.add(ws, "assistant", answer)
    
    return ChatResponse(
        answer=answer,
        sources=sources,
        links=links,
        images=images,
        followups=follow,
        default_tab="answer",
        workspace_id=ws
    )


@app.post("/api/rag", response_model=ChatResponse)
def rag_mode(req: ModeRequest):
    """
    RAG Mode - Search uploaded documents only.
    Production-level LangGraph implementation.
    """
    q = req.message.strip()
    ws = req.workspace_id
    
    memory.add(ws, "user", q)
    
    try:
        # Run the RAGOnlyGraph pipeline
        state = rag_graph.run(q, ws)
        answer = state.get("answer", "No answer generated.")
        sources = state.get("sources", [])
        follow = state.get("followups", [])
    except Exception as e:
        print(f"RAG error: {e}")
        answer = f"RAG mode encountered an error: {str(e)[:100]}"
        sources = []
        follow = []
    
    memory.add(ws, "assistant", answer)
    
    return ChatResponse(
        answer=answer,
        sources=sources,
        links=[],
        images=[],
        followups=follow,
        default_tab="answer",
        workspace_id=ws
    )


@app.post("/api/agentic", response_model=ChatResponse)
def agentic_mode(req: ModeRequest):
    """
    Agentic Mode - Multi-agent RAG with Planner, File, Web, Knowledge, Image, Synthesizer.
    Production-level LangGraph implementation.
    """
    q = req.message.strip()
    ws = req.workspace_id
    
    memory.add(ws, "user", q)
    print(f"\nπŸ€– AGENTIC MODE (LangGraph): {q}")
    
    try:
        # Run the AgenticRAGGraph pipeline
        state = agentic_graph.run(q, ws)
        answer = state.get("answer", "No answer generated.")
        sources = state.get("sources", [])
        links = state.get("links", [])
        images = state.get("images", [])
        follow = state.get("followups", [])
    except Exception as e:
        print(f"Agentic error: {e}")
        answer = f"Agentic mode encountered an error: {str(e)[:100]}"
        sources = []
        links = []
        images = []
        follow = []
    
    # Always get images if none from agentic
    if not images:
        images = tavily_images_safe(q)
    
    memory.add(ws, "assistant", answer)
    print(f"  βœ… AgenticGraph: Completed with {len(sources)} sources, {len(images)} images")
    
    return ChatResponse(
        answer=answer,
        sources=sources,
        links=links,
        images=images,
        followups=follow,
        default_tab="answer",
        workspace_id=ws
    )


# =======================================================
# PRODUCT MVP ENDPOINT - Generates MVP Blueprints
# =======================================================
class ProductMVPRequest(BaseModel):
    message: str
    workspace_id: str = "default"
    mode: str = "product_mvp"


@app.post("/api/product_mvp", response_model=ChatResponse)
def product_mvp_mode(req: ProductMVPRequest):
    """
    Product MVP Mode - Generates comprehensive MVP blueprints from product ideas.
    Includes product name, pitch, target users, features, architecture, tech stack, and more.
    """
    q = req.message.strip()
    ws = req.workspace_id
    
    memory.add(ws, "user", q)
    print(f"\nπŸš€ PRODUCT MVP MODE: {q}")
    
    # Research similar products and market
    market_research = ""
    try:
        results = search_tool.search(f"{q} startup MVP product", num_results=3)
        if results:
            for r in results:
                url = r.get("url", "")
                text = browse_tool.fetch_clean(url)
                if text:
                    market_research += text[:800] + "\n\n"
    except Exception as e:
        print(f"Market research error: {e}")
    
    prompt = f"""You are a PRODUCT BUILDER AI that creates comprehensive MVP blueprints.

The user wants to build: {q}

{f"MARKET RESEARCH (use for context):{chr(10)}{market_research}" if market_research else ""}

Generate a COMPLETE MVP Blueprint with the following sections. Use markdown formatting with tables where appropriate:

# πŸ“„ MVP Blueprint – [Product Name]
A one-line description of the product.

## 1. Product Name
Create a catchy, memorable product name.

## 2. One‑Line Pitch
A compelling pitch in quotes that explains the value proposition.

## 3. Target Users
Create a markdown table with columns: Persona | Age | Occupation | Goals | Pain Points | How [Product] Helps
Include 4-5 different user personas.

## 4. Problems to Solve
List 5 key problems the product solves with bullet points.

## 5. MVP Features
Create a table with: Feature | Description | Priority (Must-have/Nice-to-have)
Include 8-10 features.

## 6. User Journey (Step‑by‑Step)
Number each step of the user journey from landing to retention.

## 7. System Architecture
Create an ASCII diagram showing the system components and their connections.
Include: Frontend, Backend, Database, APIs, Third-party services.

## 8. Database Tables
Create a table showing the main database tables with: Table | Columns | Notes

## 9. API Endpoints (REST)
Create a table with: Method | Endpoint | Description | Auth Required

## 10. Tech Stack
Create a table with: Layer | Technology | Reason
Cover: Frontend, Backend, Auth, Database, Cache, Storage, Hosting, CI/CD, Monitoring

## 11. Future Features (Post‑MVP)
List 8 features for after MVP launch.

## Next Steps
List 5 actionable next steps to start building.

End with: **Happy building! πŸš€**

Be detailed, practical, and use real-world best practices. Make it production-ready."""

    msgs = build_context(ws, prompt)
    answer = llm.invoke(msgs).content
    
    # Generate follow-up questions
    follow = [
        "Generate wireframes for core screens",
        "Create a development timeline",
        "Estimate the MVP budget",
        "Design the database schema in detail",
        "Write user stories for MVP features"
    ]
    
    memory.add(ws, "assistant", answer)
    print(f"  βœ… Product MVP: Blueprint generated")
    
    return ChatResponse(
        answer=answer,
        sources=[],
        links=[],
        images=[],
        followups=follow,
        default_tab="answer",
        workspace_id=ws
    )


# =======================================================
# VIDEO BRAIN ENDPOINT - YouTube Video Analysis with Transcript + Web Fallback
# =======================================================
from tools.youtube_tool import YouTubeTool
youtube_tool = YouTubeTool()

# Store video transcripts in memory per workspace
video_transcripts = {}

class VideoBrainRequest(BaseModel):
    message: str
    workspace_id: str = "default"
    mode: str = "video_brain"
    youtube_url: str = ""
    transcript: str = ""  # User can paste transcript directly


@app.post("/api/video_brain", response_model=ChatResponse)
def video_brain_mode(req: VideoBrainRequest):
    """
    Video Brain Mode - Analyzes YouTube videos.
    Uses user-pasted transcript (priority), then tries auto-extraction, then web fallback.
    """
    q = req.message.strip()
    ws = req.workspace_id
    youtube_url = req.youtube_url
    user_transcript = req.transcript.strip() if req.transcript else ""
    
    memory.add(ws, "user", q)
    print(f"\nπŸŽ₯ VIDEO BRAIN MODE: {q}")
    print(f"  πŸ“Ί YouTube URL: {youtube_url}")
    print(f"  πŸ“ User transcript provided: {len(user_transcript)} chars")
    
    # Allow transcript-only mode (no URL required if user pastes transcript)
    if not youtube_url and not user_transcript:
        return ChatResponse(
            answer="⚠️ Please provide a YouTube URL or paste the transcript directly.",
            sources=[],
            links=[],
            images=[],
            followups=[],
            default_tab="answer",
            workspace_id=ws
        )
    
    # Handle transcript-only mode
    if youtube_url == "user_provided_transcript" or not youtube_url:
        youtube_url = "user_provided_transcript"
        video_id = "user_transcript"
    else:
        video_id = youtube_tool.extract_video_id(youtube_url)
    
    cache_key = f"{ws}_{video_id}"
    
    # PRIORITY 1: User-pasted transcript
    transcript_text = ""
    transcript_data = None
    
    if user_transcript:
        transcript_text = user_transcript[:15000]  # Allow longer user transcripts
        print(f"  βœ… Using user-pasted transcript: {len(transcript_text)} chars")
        # Cache it for follow-up questions
        video_transcripts[cache_key] = {"success": True, "transcript": transcript_text, "user_provided": True}
    
    # PRIORITY 2: Check cache
    if not transcript_text and cache_key in video_transcripts:
        transcript_data = video_transcripts[cache_key]
        if transcript_data.get("success"):
            transcript_text = transcript_data.get("transcript", "")[:8000]
            print(f"  πŸ“‹ Using cached transcript")
    
    # PRIORITY 3: Try auto-extraction
    if not transcript_text:
        print(f"  πŸ”„ Attempting auto-extraction for video: {video_id}")
        transcript_data = youtube_tool.get_transcript(youtube_url)
        
        if transcript_data.get("success"):
            video_transcripts[cache_key] = transcript_data
            transcript_text = transcript_data.get("transcript", "")[:8000]
            print(f"  βœ… Auto-extracted transcript: {len(transcript_text)} chars")
        else:
            print(f"  ⚠️ Auto-extraction failed: {transcript_data.get('error')}")
    
    # If no transcript, use web search fallback
    video_context = ""
    sources = []
    links = []
    
    if not transcript_text:
        print(f"  🌐 Using web search fallback...")
        try:
            # Search for video info and summaries
            if search_tool:
                search_queries = [
                    f"youtube video {video_id} summary transcript",
                    f"youtube {video_id} key points explained"
                ]
                
                for sq in search_queries[:1]:  # Just one search to save time
                    results = search_tool.search(sq, num_results=4)
                    
                    # Get Tavily AI answer
                    if results and results[0].get("tavily_answer"):
                        video_context += f"[Video Summary]: {results[0]['tavily_answer']}\n\n"
                    
                    for r in results:
                        url = r.get("url", "")
                        title = r.get("title", "")
                        content = r.get("content", "")
                        
                        if content:
                            video_context += f"[{title}]: {content[:1000]}\n\n"
                            links.append({"title": title, "url": url, "snippet": content[:150]})
                            sources.append({"title": title, "url": url})
                
                print(f"  πŸ“„ Web fallback gathered: {len(video_context)} chars, {len(sources)} sources")
        except Exception as e:
            print(f"  ❌ Web search fallback error: {e}")
    
    # Build prompt
    q_lower = q.lower()
    is_summary = any(word in q_lower for word in ["summarize", "summary", "overview", "main points", "key takeaways", "what is this about"])
    
    if transcript_text:
        # Have real transcript
        if is_summary:
            prompt = f"""You are VIDEO BRAIN AI - an expert at analyzing YouTube videos.

VIDEO TRANSCRIPT (with timestamps [MM:SS]):
{transcript_text}

USER REQUEST: {q}

Provide a comprehensive summary with:
1. **Overview**: One paragraph describing what the video is about
2. **Key Points**: 5-7 main takeaways with timestamps
3. **Important Details**: Any specific facts, figures, or examples mentioned
4. **Actionable Insights**: What viewers should do or remember

Use the actual content from the transcript. Reference timestamps like [5:30] when citing specific parts."""
        else:
            prompt = f"""You are VIDEO BRAIN AI - an expert at analyzing YouTube videos.

VIDEO TRANSCRIPT (with timestamps [MM:SS]):
{transcript_text}

USER QUESTION: {q}

Answer the question using ONLY the information from the transcript above.
- Be specific and cite timestamps when relevant
- If the answer is not in the transcript, say so honestly
- Format your response clearly with bullet points if appropriate"""
        
        sources = [{"title": "πŸŽ₯ YouTube Video (Transcript)", "url": youtube_url}]
        links = [{"title": "Source Video", "url": youtube_url, "snippet": f"Video ID: {video_id} - Full transcript available"}]
        
    elif video_context:
        # Have web search fallback context
        prompt = f"""You are VIDEO BRAIN AI. I couldn't get the direct transcript, but found related information about this video.

VIDEO URL: {youtube_url}
VIDEO ID: {video_id}

AVAILABLE INFORMATION FROM WEB:
{video_context[:6000]}

USER QUESTION: {q}

Based on the available information:
1. Answer the user's question as best as you can
2. Be clear that this is based on web search results, not the actual transcript
3. If summarizing, provide the key points found
4. Suggest the user can paste the transcript directly for more accurate analysis"""
    else:
        # No information available
        error_msg = transcript_data.get("error", "Unknown error") if transcript_data else "Could not fetch transcript"
        prompt = f"""I couldn't analyze the YouTube video.

Video URL: {youtube_url}
Error: {error_msg}
User Question: {q}

Please explain:
1. Why the transcript couldn't be fetched (network/DNS issues on this server)
2. Alternative: The user can:
   - Open YouTube, click "..." under the video, select "Show transcript"
   - Copy and paste the transcript text here
   - I can then analyze it accurately
3. Or they can try a different video"""
    
    try:
        msgs = build_context(ws, prompt)
        answer = llm.invoke(msgs).content
    except Exception as e:
        print(f"  ❌ LLM error: {e}")
        answer = f"Error generating response: {str(e)[:100]}"
    
    # Follow-up questions
    if transcript_text or video_context:
        follow = [
            "What are the main arguments or points made?",
            "Summarize this in 3 bullet points",
            "What examples or case studies are mentioned?",
            "What should I learn from this video?",
            "Explain the most complex concept in simple terms"
        ]
    else:
        follow = [
            "Paste the transcript text here",
            "Try a different YouTube video",
            "How do I get a YouTube transcript?"
        ]
    
    # Add video source if not already added
    if not sources:
        sources = [{"title": "πŸŽ₯ YouTube Video", "url": youtube_url}]
    if not links:
        links = [{"title": "Source Video", "url": youtube_url, "snippet": f"Video ID: {video_id}"}]
    
    memory.add(ws, "assistant", answer)
    print(f"  βœ… Video Brain: Response generated")
    
    return ChatResponse(
        answer=answer,
        sources=sources,
        links=links,
        images=[],
        followups=follow,
        default_tab="answer",
        workspace_id=ws
    )