v1.1.0

v1.1.0 (2025-09-28)

This release introduces hybrid entity retrieval, entity extraction and tracking, token management improvements, performance optimizations, and several UI/readability fixes.

Docker image

The Docker image for this release is published to GitHub Container Registry: ghcr.io/florentb974/graphrag:v1.1.0

Highlights

Hybrid entity retrieval and extraction

• Add entity extraction pipeline with background processing and status updates.
• Implement entity normalization and deduplication to improve result quality.
• Add entity ID retrieval from names and improve logging for retrieval processes.

Retrieval improvements

• Introduce hybrid retrieval mode with reduced expand depth for performance.
• Add a minimum similarity threshold to filter low-quality chunks.
• Enhance retrieval logic to exclude low-similarity chunks from response generation.

Token management and LLM handling

• Ensure model type conversion to string in OpenAI response generation.
• Implement safe fallbacks for truncated responses and output token management.
• Parallelize LLM calls for entity extraction where applicable.

Performance and reliability

• Parallelize embedding generation for entity mode.
• Add embedding validation and exponential backoff.
• Reduce expand depth in hybrid retrieval mode to improve performance.

UI and UX improvements

• Limit displayed entities and chunks to 10 for improved UI readability.
• Fix source display and improve graph readability.
• Implement status updates for entity extraction operations in the UI.

Misc and housekeeping

• Several code cleanups and logging enhancements.
• Merge updates from main and minor fixes.

Commits since v1.0.0

• feat: Hybrid entity retrieval and extraction improvements (#12)

(See full commit history for details.)

---

If you spot issues in this release, please open an issue