v0.25.0
TabbySummary
This release introduces significant new capabilities for the Answer Engine, including exposing its thinking process and enabling repository file access. Users will also benefit from improved stability with automatic embedding retries and an enhanced user interface experience.
New Features
- Exposed the thinking process of the Answer Engine within thread messages.
- Enabled the Answer Engine to access the repository's directory file list as needed.
- Enabled the use of the "@" symbol to mention a symbol in the Chat Sidebar.
- Provided repository-aware default question recommendations on the Answer Engine.
Bug Fixes
- Resolved the deserialization issue related to "finish_reason" when receiving chat responses from the LiteLLM Proxy Server.
Improvements
- Increased the default parallelism from 1 to 4, which may increase VRAM usage.
- Introduced a new embedding kind `llama.cpp/before_b4356_embedding` for llamafile or other embedding services using the legacy llama.cpp embedding API.
- Provided a configuration option to truncate text content before dispatching it to the embedding service.
- Bumped the llama.cpp version to b4651.
- Enabled automatic retries for embedding when the service occasionally fails due to llama.cpp issues.
- Enhanced the user interface experience for the Answer Engine.