Conversation
Bump ai SDK to 6.0.145, @google/genai to 1.48.0, @biomejs/biome to 2.4.10, and various other packages to latest minor/patch versions. Add gpt-tokenizer.
Add a "Test Search" button that opens a dialog to test semantic similarity queries against indexed lorebook entries. Shows activated entries above threshold and dim entries below. Also fix indexing status query to count all entries (not just enabled) and handle null similarity_threshold.
Add three new agent workflow nodes: getLorebook, searchLorebook, and addLorebookEntry. Register their input/output handles and wire up multi-output support for search and add nodes in the runner.
Add ResolvedParameters type and reportResolvedParams callback to AIEvent. Capture the actual parameters sent to the provider and display them in the ParametersTab alongside the original input parameters.
Use gpt-tokenizer's encode() for accurate token counting instead of the custom word/punctuation heuristic that had ~8% error margin.
Extract INFERENCE_TIMEOUT_MS constant (5 min) for local model support. Fix modelsStore to pass profile_id when refreshing after create/update. Fix EmbeddingModel type parameter deprecation.
- Added Node.js engine requirement (>=24.15.0) in package.json. - Updated various package versions in package.json and pnpm-lock.yaml: - @tauri-apps/plugin-dialog: ~2.7.0 - @tauri-apps/plugin-fs: ~2.5.0 - immer: ^11.1.4 - react-resizable-panels: ^4.10.0 - uuid: ^14.0.0 - @vitejs/plugin-react: ^6.0.1 - vite: ^8.0.10 - typescript: ~6.0.3 - Adjusted tsconfig.json path configuration for better compatibility. - Updated code-quality workflow to use Node.js LTS version.
…of NarratrixAI features and capabilities
…ge with new filtering and layout features
Comment out the registration in registry/grid the same way memory is, and make GridSidebar tolerate saved layouts that reference unregistered ids.
Adds a source toggle (From Character / Pick Lorebook) so workflows can output a fixed lorebook without going through a character.
- Group participants by type with counts on filter tabs - Show description as subtitle and add agent badge overlay on avatar - Match popover backgrounds and tighten side offset - Refresh display settings with icon tiles and per-toggle descriptions
Opus 4.7 rejects thinking.type "enabled" and requires "adaptive" with an effort level instead of budgetTokens. Detect 4.7+ / 5+ models by name and emit the right shape per provider; older models keep the existing enabled+budget behavior.
Removes the redundant theme picker from the settings page and wires the sidebar's theme switcher to write through to the active profile's appearance.theme, so the choice survives across sessions instead of getting reverted by SettingsPage's profile sync on mount.
- Skip entries with mismatched vector dimensions during retrieval and in the test-search dialog so a model swap can't crash inference or silently produce garbage similarity scores. - Clear stored vectors automatically when a lorebook's embedding model changes, matching what the form already promised the user. - Strip embedding_model_id and vector_content on V1 lorebook import to prevent a foreign profile's encrypted API key from being reused. - Pass tauriFetch to the AWS Bedrock embedding factory and throw clear errors when API keys or required config are missing instead of forwarding "None" placeholders to the provider. - Remove success toasts for indexing actions per project policy and surface previously console-only errors via toasts in the lorebook form, entry dialog, drag-reorder, toggle, and delete flows. - Re-throw from clearIndex so the caller's catch fires instead of reporting false success. - Clear stale vectors when an entry's content is emptied, log parseStoredVector failures, and gate Index All on a configured model.
Keyword matching now stacks with semantic similarity instead of being mutually exclusive. When RAG is enabled, a keyword hit adds a +0.5 boost to cosine similarity (capped at 1) so authored triggers can activate entries that semantic search alone would miss. Empty keywords opt out naturally. - Always show keyword fields (Keywords, Trigger Chance, Case Sensitive, Match Partial Words) in the entry dialog regardless of rag_enabled. - Always show the Keywords column in the entries table; Indexed column is now an addition rather than a replacement. - Apply trigger_chance at runtime (was wired into the form but never consulted by the matcher). - Rework Test RAG Search dialog: Enter to submit, autofocus, threshold preview slider, recent-query chips, search timing, score bars with threshold marker, click-to-expand results, keyword-hit "K" badge, inline stats bar.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds semantic search (RAG) capabilities to the Lorebook system, along with supporting infrastructure for embedding models.
Embedding Models Support
embedding-serviceandlorebook-indexing-servicefor vectorizing lorebook entriesLorebook RAG Integration
apply-lorebookformatter updated to support both keyword and semantic search modesRAG Test Dialog
Lorebook Agent Tool Nodes
getLorebook,searchLorebook,addLorebookEntryConsole Inspector: Resolved Parameters
ResolvedParameterstype showing the actual values sent to the AI providerOther Improvements
gpt-tokenizerfor accurate countingmodelsStoreto passprofile_idwhen refreshing after create/updateTest plan