feat: Recall the original text of the knowledge base content rather than retrieving the generated memory. (#1057)

* update reader and search strategy

* set strategy reader and search config

* fix install problem

* fix

* fix test

* turn off graph recall

* turn off graph recall

* turn off graph recall

* fix Searcher input bug

* fix Searcher

* fix Search

* fix bug

* adjust strategy reader

* adjust strategy reader

* adjust search config input

* reformat code

* re pr

* format repair

* fix time issue

* develop feedback process

* feedback handler configuration

* upgrade feedback using

* add threshold

* update prompt

* update prompt

* fix handler

* add feedback scheduler

* add handler change node update

* add handler change node update

* add handler change node update

* add handler change node update

* fix interface input

* add chunk and ratio filter

* update stopwords

* fix messages queue

* add seach_by_keywords_LIKE

* add doc filter

* add retrieve query

* add retrieve queies

* patch info filter

* add log and make embedding safety net

* add log and make embedding safety net

* deduplicate add objects

* use _add_memories_parallel

* delete Special characters

* delete Special characters

* delete Special characters

* delete Special characters

* add source_doc_id

* add source_doc_id

* add reranker in init com..

* fix circle import

* add feedback judgement

* add feedback judgement

* add pref feedback

* add pref feedback

* patch: get_memory func filter user id and make page chunk

* add total num

* add total num

* add milvus pagination

* fix merge implicit explicit pref

* fix merge implicit explicit pref

* fix merge implicit explicit pref

* fix json load bug

* knowledge raw_text replace memory

* knowledge raw_text replace memory

* knowledge raw_text replace memory

* unuse rerank

* backtrack knowledge retrieval

* norerank knowledge data

* use embedding rerank knowledge

* add chunk memories

* feat: mix search upload file raw content

* feedback soures set empty

* chat upload file can be chunk

* fix: when rerank query is too long

* fix too long rerank

* fix too long rerank

* make mem_reader config

* make mem_reader config

* transmit search_memory_type param

* refactor format

* add file url to knowledge db

* add log

* fix

* search neighbor_discovery

* record fallback ratio

* feedback do not change memories with edges

* add skill

* fix test

---------

Co-authored-by: 黑布林 <11641432+heiheiyouyou@user.noreply.gitee.com>
Co-authored-by: CaralHsi <caralhsi@gmail.com>
Co-authored-by: chunyu li <78344051+fridayL@users.noreply.github.com>
Co-authored-by: Dubberman <48425266+whipser030@users.noreply.github.com>

Author: 5 people

src/memos/mem_scheduler/task_schedule_modules/handlers/mem_read_handler.py

@@ -134,6 +134,7 @@ def _process_memories_with_reader(
                 )
                 return
 
+            # Get the original fast memory (raw chunk) items
             memory_items = []
             for mem_id in mem_ids:
                 try: