Skip to content

Releases: hensing/searchless-ngx

v.0.1.4 - gemini chunking fix

06 Mar 21:08
@hensing hensing
v0.1.4
41b025c

Choose a tag to compare

View all tags
v.0.1.4 - gemini chunking fix Latest
Latest

Release Notes - v0.1.4 🛡️
This release focuses on Stability and Large Document Support, ensuring that Searchless-ngx can handle even the most extensive document libraries without hitting API limits.

Key Improvements in v0.1.4
📦 Smart Embedding Batching
The Gemini Embedding API has a hard limit of 100 requests per batch. Searchless-ngx now automatically detects large documents and splits them into compliant batches. This prevents the INVALID_ARGUMENT errors previously encountered with documents exceeding 100 chunks.

✂️ Configurable Document Truncation
To prevent resource exhaustion and ensure high-quality search results, we've introduced a configurable chunk limit:

MAX_CHUNKS_PER_DOC: New environment variable (Default: 100).
Capacity: 100 chunks cover approximately 25 DIN-A4 pages of text.
Graceful Handling: Documents exceeding this limit are truncated at the end, and a warning is logged.
🧪 Enhanced Reliability
Automated Batching Tests: We've added a new mocked test suite to verify the batching logic without requiring internet access.
Expanded Test Coverage: This release is verified by 18 automated tests, ensuring core stability.

Assets 2
Loading

v0.1.3 Initial Release

06 Mar 20:44
@hensing hensing
v0.1.3
e22e164

Choose a tag to compare

View all tags
v0.1.3 Initial Release

Release Notes - v0.1.3 🚀

We are proud to announce the Initial Release of Searchless-ngx (v0.1.3), the first production-ready version of our Agentic RAG MCP Server for Paperless-ngx.

What is Searchless-ngx?

Searchless-ngx transforms your Paperless-ngx instance from a static archive into an intelligent, conversational agent. By leveraging the Model Context Protocol (MCP) and Agentic RAG, it allows modern LLMs (like Gemini or GPT-4) to natively search, filter, and reason over your personal documents.

Key Features in this Release

🎴 Professional Markdown "Cards"

Search results are now presented as high-quality interactive cards designed specifically for Open WebUI:

  • Linked Headers: Document titles link directly to the document detail view in Paperless.
  • Dynamic Deep-Linking: Correspondent names and Tags are now clickable links that filter for related documents.
  • Structural Preservation: OCR snippets now respect original line breaks and paragraph structure.
  • Concise Layout: Strict 7-line snippet limits keep your chat clean and readable.

🔍 Advanced Hybrid Search

  • Exact Metadata API: Leverage the full power of Paperless-ngx filtering (correspondents, tags, dates).
  • Semantic Vector Search: Use ChromaDB and Gemini embeddings to find documents by meaning (e.g., "Find food receipts from Berlin").
  • Custom Field Visibility: Custom fields are beautifully integrated and resolved into readable names.

🏗️ Enterprise-Grade Reliability

  • Paginated Cache: Efficiently handles large libraries with hundreds of tags and correspondents.
  • Strict JSON Schema: 100% compatible with strict MCP parsers (no anyOf or null types).
  • Comprehensive Verification: This release is verified by a full suite of 17 automated tests covering all core components.

Installation

For detailed installation and Open WebUI setup instructions, please refer to the README.md and WEBUI_SETUP.md.

Thank you for choosing Searchless-ngx!

Loading