Skip to content

Claude/reverent bose#156

Closed
addy-2709genius wants to merge 57 commits intoendee-io:masterfrom
addy-2709genius:claude/reverent-bose
Closed

Claude/reverent bose#156
addy-2709genius wants to merge 57 commits intoendee-io:masterfrom
addy-2709genius:claude/reverent-bose

Conversation

@addy-2709genius
Copy link
Copy Markdown

No description provided.

shaleenji and others added 30 commits January 25, 2026 07:36
Update install.sh
…registry image. (endee-io#6)

Co-authored-by: Burhan Kapdawala <burhan@endee.io>
…er_hybrid_correction

Fix: Delete vectors from sparse storage when deleting by filter
feat: Implement backup download and upload API endpoints
quantization optimization and bug fixes
clean of DISABLE_HYBRID_QUANTIZATION and fixing of binary index bug
…dee-io#30)

* first commit

* readme update

---------

Co-authored-by: Shaleen Garg <shaleengarg.in@gmail.com>
* filter aware search

* multiple entry points to base layer

* progressive filtering approach

* budget based pruning

* filter_params
* fix:sparse storage db path correction

* Apply suggestion from @Copilot

change variable name

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

---------

Co-authored-by: rajeshkomaravelli <rajeshkomuravelli333@gmail.com>
Co-authored-by: Vineet Dwivedi <164136199+vindwid@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
* backfill
* vector cache - dynamic hot cache
…_api

add meta and filter in application/json in insert api
Co-authored-by: rajeshkomaravelli <rajeshkomuravelli333@gmail.com>
shaleenji and others added 27 commits February 25, 2026 12:51
This has major changes related to graph, backfill, vector cache and SIMD and few bug fixes. Not an ideal commit.
1. Backfill now does not require extra neighbors. This means that the index will be consistent
2. SIMD optimization for AVX512 and SVE2
3. Narrow beam search at L1 is replaced with greedy search
4. Hybrid quantization is removed. It had memory efficient and speed gain at the cost of minor impact on recall.
5. We are not multiplying in batches. This will make porting to GPU easier. We can also optimize the registers with sticky query vector.
6. Vector cache has 
7. Bug fix where enry id was updated. It was creating a separate graph.
9. Bug fix when updating for vector cache refresh.
* bug fix

* memory order

* explicit checks
* async backup job

* streaming backup downloads

* backup-system docs

* using only operation_mutex

* correction

* Stream downloading backup updates (With frontend) (endee-io#40)

* testing

* feat: add download token for backups

* refactor: remove authToken from generate_download_key

* chore: update web ui link

* refactor: get authToken as query param to verify download

* refactor: use json_errors for download backup api

* refactor: add version variable for frontend

* chore: update UI version

* only using unordered_map for tarcking

* Using Backupstore as separate utility/helper class that handles file-level operations (tar, json, paths, tracking)

* backup_store file moved to storage folder
Co-authored-by: rajeshkomaravelli <rajeshkomuravelli333@gmail.com>
* Acquire per-index operation_mutex in deleteIndex to prevent race with concurrent writes

* Acquire per-index operation_mutex in deleteIndex
Co-authored-by: Shaleen Garg <shaleengarg.in@gmail.com>
* fix:sve2 instruction compile error

* changing from size_t to svptrue

---------

Co-authored-by: rajeshkomaravelli <rajeshkomuravelli333@gmail.com>
Co-authored-by: Pankaj Singh <pankajsingh1397@gmail.com>
Co-authored-by: rajeshkomaravelli <rajeshkomuravelli333@gmail.com>
Co-authored-by: Pankaj Singh <pankajsingh1397@gmail.com>
Co-authored-by: Pankaj Singh <pankajsingh1397@gmail.com>
Co-authored-by: Pankaj Singh <pankajsingh1397@gmail.com>
Co-authored-by: Pankaj Singh <pankajsingh1397@gmail.com>
Co-authored-by: Pankaj Singh <pankajsingh1397@gmail.com>
- Replace word-based chunking with recursive character chunking in
  rag_app/utils.py; tries paragraph/sentence separators before splitting
  mid-sentence, with character-level fallback
- Update CHUNK_SIZE (500 words → 1000 chars) and CHUNK_OVERLAP
  (50 words → 200 chars) in config.py to match character-based logic
- Add .streamlit/config.toml forcing light theme base so the app
  renders correctly regardless of system dark mode
- Strengthen app.py CSS to override Streamlit dark theme variables
- Replace default chat avatars with DiceBear SVG icons for user and
  assistant

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.