Version v0.1.111¶

This release moves the agent's ReAct loop from the backend to the frontend, unifying it with slash commands for a more streamlined and consistent user experience. The backend now acts as a stateless LLM proxy, while tools are auto-generated and dispatched through a common feature handler. Several enhancements improve tool execution, search enrichment, and user interface elements like the agent log drawer.

New Features¶

Moved the agent ReAct loop to the frontend, unifying tool dispatch with slash commands to eliminate display-model drift (818dd3) (Eric Ma)
Added a respond tool that synthesizes AI nodes linked to context nodes (818dd3) (Eric Ma)
Introduced a code tool with auto-execution and output capture capabilities (818dd3) (Eric Ma)
Enhanced search enrichment by automatically fetching the top 2 pages per search (818dd3) (Eric Ma)
Implemented thinking-token stripping to clean up agent responses by removing <think> tags and markdown fences (818dd3) (Eric Ma)
Enabled canvas context for replies, including parent node content and code (818dd3) (Eric Ma)
Added an agent log side drawer with a copy button for easier access and sharing of logs (818dd3) (Eric Ma)
Improved reliability with rate-limit retry using 3x progressive backoff (818dd3) (Eric Ma)
Restricted edge creation to only via createLinkedNode to prevent bidirectional edges (818dd3) (Eric Ma)
Increased the maximum number of agent tool calls to 30 (818dd3) (Eric Ma)

Bug Fixes¶

No bug fixes included in this release.

Deprecations¶

No deprecations in this release.