Starstruk Logo MarkStarstruk

1.0.2

sarperavci/CloudflareBypassForScraping1.0.2Apr 22, 2026by Dsantra92

AI Summary

Version 1.0.2 focuses on improving code search quality through parallelized parsing using ColBERT and BM25 embeddings, and introduces new multimodal capabilities for handling attachments and documents.

Key Highlights

  • Parallelized parsing with ColBERT and BM25 embeddings.
  • New parsing module using tree-sitter and Qdrant for vector indexing.
  • Enhanced multimodal support for document attachments.
  • Parallelized parsing improves graph creation speed.

New Features

  • New `app/src/parsing/` module with Python bindings.
  • Qdrant Hybrid Search (dense and sparse vector indexing).
  • Attachment handling updates including retroactive application.
  • Multimodal processing now supports document types.
  • New `update_agent` endpoint for dynamic agent switching.

Full Release Notes

Parallelized parsing with ColBERT and BM25 embeddings
## What's Changed

We have parallelized the parsing process and packaged the entire parsing into one sub package for parsing.So the old `create_graph_rs` Rust module has been replaced with a new app/src/parsing/ module with ColBERT/BM25 hybrid vector indexing via Qdrant for superior code search quality.

### Core Infrastructure
- New Parsing module (app/src/parsing/): Complete rewrite with tree-sitter for multi-language support, 800+ lines of tag extraction, and Python bindings via PyO3. Moved parsing queries to the new module structure.
- Qdrant Hybrid Search: Added dense (ColBERT) + sparse (BM25) vector indexing to the code graph pipeline. Non-fatal indexing failures don't block graph creation. Includes deletion handling on project cleanup.
- Qdrant configuration: Added Qdrant configuration support (host, port, gRPC port, API key, HTTPS) alongside existing Neo4j config.

### Multimodal & Conversation Improvements
- Attachment Handling: Extended conversation creation to accept attachment_ids. First-message attachments are now stored in Redis and applied retroactively if the client redirects without re-sending them.
- Document Attachments: Multimodal processing now handles both images and documents (AttachmentType.DOCUMENT), not just images.
- Agent Updates: Added update_agent endpoint to conversation controller for dynamic agent switching.

## PR's merged
* mode switch backend API added with update in DB by @shmbhvi101 in https://github.com/potpie-ai/potpie/pull/728
* Fix rust build for docker by @Dsantra92 in https://github.com/potpie-ai/potpie/pull/744
* fix(auth): send Slack alerts for email/password signups by @yashkrishan in https://github.com/potpie-ai/potpie/pull/747
* unbound local error while parsing using priority 3 github token by @BrhKmr23 in https://github.com/potpie-ai/potpie/pull/748
* Add faster parsing + colbert embeddings by @Dsantra92 in https://github.com/potpie-ai/potpie/pull/746

## New Contributors
* @BrhKmr23 made their first contribution in https://github.com/potpie-ai/potpie/pull/748

**Full Changelog**: https://github.com/potpie-ai/potpie/compare/v1.0.1...1.0.2