Skip to content

RELEASE_NOTES.md

Latest commit

 

History

History
77 lines (54 loc) · 4.64 KB

RELEASE_NOTES.md

File metadata and controls

77 lines (54 loc) · 4.64 KB

Release Notes - v0.1.5 🧭

This release introduces Search Resilience & Smart Fallback Strategies, making the agent much more proactive when initial search results are empty.

Key Improvements in v0.1.5

🧭 Smart Search Fallbacks

The search tools now include explicit "Retry Logic" for the LLM. If a narrow search (e.g., within a specific date range) yields no results, the agent is now instructed to:

  • Relax Filters: Automatically try broader criteria, such as removing date boundaries or broadening the query.
  • Contextual Retries: Inform the user if results are found in a different timeframe than requested.

💬 Descriptive Empty-Result Feedback

Search results are no longer a silent "No results". The tools now return detailed context:

  • Applied Filters: Lists all parameters (IDs, dates, query) that were used, helping the LLM reason about its next search attempt.
  • Clear Guidance: Provides direct recommendations to the LLM on how to broaden the search.

🆔 Strict Metadata ID Enforcement

To prevent the LLM from "hallucinating" or guessing correspondent/tag IDs:

  • Mandatory Lookup: The get_paperless_master_data tool now explicitly warns against guessing IDs and requires using it for lookup first.

🧪 Verified Stability

  • Raw Protocol Checks: This release has been verified using the raw MCP protocol checker.
  • Test Suite: Verified by the full suite of 18 automated tests.

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.

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 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!