Configuration
All settings are accessible from the ⚙ Settings tab in the Dashboard or via VS Code’s settings JSON.
General
| Setting | Default | Description |
contextManager.showStatusBar | true | Show active project in the status bar |
contextManager.confirmDelete | true | Confirmation dialog before deleting items |
contextManager.maxKnowledgeCardsInContext | 10 | Maximum cards to include in AI prompts (1–20) |
Auto-Capture
| Setting | Default | Description |
contextManager.autoCapture.enabled | true | Enable automatic observation recording |
contextManager.autoCapture.learnFromAllParticipants | true | Learn from all chat participant interactions |
contextManager.autoCapture.maxObservations | 50 | Max observations in circular buffer |
Auto-Distill
| Setting | Default | Description |
contextManager.autoDistill.enabled | true | Automatically distill observations into conventions and knowledge cards at compaction checkpoints |
contextManager.autoDistill.intervalMinutes | 30 | Minimum minutes between automatic distillation runs per project |
contextManager.autoDistill.dedupThreshold | 0.8 | Jaccard similarity threshold for duplicate detection (0.5–1.0) |
Project Intelligence
| Setting | Default | Description |
contextManager.intelligence.enableTieredInjection | true | Auto-inject conventions and top tool hints into every chat prompt (Tier 1+2) |
contextManager.intelligence.enableStalenessTracking | true | File-based staleness tracking on working notes and knowledge cards |
contextManager.intelligence.stalenessAgeDays | 30 | Days before knowledge cards are flagged as age-stale (7–365) |
contextManager.intelligence.tier1MaxTokens | 400 | Token budget for Tier 1 (always-injected) learnings (100–1000) |
contextManager.intelligence.tier2MaxTokens | 400 | Token budget for Tier 2 (task-relevant) learnings (100–1000) |
AI Model Routing
| Setting | Default | Description |
contextManager.intelligence.autoLearn.modelFamily | (auto) | Preferred model family for background auto-learn extraction. Use this to keep continuous extraction on a smaller or cheaper model. |
contextManager.workflows.modelFamily | (auto) | Preferred model family for AI workflow actions. Template-only workflow actions ignore this setting. |
contextManager.knowledgeCards.synthesisModelFamily | (auto) | Preferred model family for AI Draft / Synthesize Card in the dashboard editor. |
Auto-Learn
| Setting | Default | Description |
contextManager.intelligence.autoLearn | true | Enable the auto-learning pipeline |
contextManager.intelligence.autoLearn.useLLM | true | Use lightweight LLM call for extraction instead of regex-only |
contextManager.intelligence.autoLearn.modelFamily | (auto) | Preferred model family for auto-learn LLM extraction (dynamic dropdown in dashboard) |
contextManager.intelligence.autoLearn.showInChat | true | Show auto-learn results as a notification toast |
contextManager.intelligence.autoLearn.extractConventions | true | Extract conventions from response text |
contextManager.intelligence.autoLearn.extractToolHints | true | Extract tool hints from search fail→success patterns |
contextManager.intelligence.autoLearn.extractWorkingNotes | true | Extract working notes from file co-access patterns |
contextManager.intelligence.autoLearn.conventionsPerRun | 1 | Max conventions to extract per interaction |
contextManager.intelligence.autoLearn.hintsPerRun | 3 | Max tool hints to extract per interaction |
contextManager.intelligence.autoLearn.notesPerRun | 2 | Max working notes to extract per interaction |
contextManager.intelligence.autoLearn.maxConventions | 15 | Max inferred conventions per project |
contextManager.intelligence.autoLearn.maxToolHints | 20 | Max tool hints per project |
contextManager.intelligence.autoLearn.maxWorkingNotes | 30 | Max inferred working notes per project |
contextManager.intelligence.autoLearn.expiryDays | 0 | Auto-expire inferred learnings after N days (0 = disabled) |
contextManager.intelligence.autoLearn.discardThreshold | 5 | Suppress a category after N discards (0 = disabled) |
Intelligence Injection
| Setting | Default | Description |
contextManager.intelligence.injectConventions | true | Include conventions in auto-injected context |
contextManager.intelligence.injectToolHints | true | Include tool hints in auto-injected context |
contextManager.intelligence.injectWorkingNotes | true | Include working notes in auto-injected context |
contextManager.intelligence.injectKnowledgeCards | true | Include knowledge cards in auto-injected context |
Card Queue
| Setting | Default | Description |
contextManager.cardQueue.enabled | true | Enable automatic response queuing |
contextManager.cardQueue.minResponseLength | 300 | Min response length to queue (50–5000) |
contextManager.cardQueue.distillBatchSize | 2 | Candidates processed per LLM call during distillation |
contextManager.cardQueue.maxCardsPerDistill | 12 | Max knowledge cards to extract per distill run |
contextManager.cardQueue.maxSize | 30 | Max candidates in the review queue |
Session Tracking
| Setting | Default | Description |
contextManager.sessionTracking.enabled | true | Track chat sessions in the Sessions tab. When disabled, no sessions are recorded and the Sessions tab stays empty. |
Hooks
| Setting | Default | Description |
contextManager.hooks.sessionStart | true | Inject selected cards and custom instructions into every prompt via session-context.txt |
contextManager.hooks.postToolUse | false | Capture tool use events as observations. Warning: creates an observation for every single tool call — very noisy. |
contextManager.hooks.preCompact | true | Capture session state before VS Code compacts conversation history |
contextManager.hooks.stop | true | Capture final exchange for card queue processing |
Prompt Customization
Customize the system prompts used by the distill pipelines. Leave empty to use defaults.
| Setting | Default | Description |
contextManager.prompts.globalInstructions | (empty) | Global custom instructions appended to all prompts |
contextManager.prompts.distillObservations | (built-in) | System prompt for distilling observations into conventions, tool hints, and working notes |
contextManager.prompts.distillQueue | (built-in) | System prompt for synthesizing card queue candidates into knowledge cards |
contextManager.prompts.synthesizeCard | (built-in) | System prompt for the dashboard’s “Synthesize Card” action |
Search
| Setting | Default | Description |
contextManager.search.enableFTS | true | Enable BM25 full-text search |
contextManager.search.maxCardResults | 5 | Max results for card search |
contextManager.search.maxSearchResults | 10 | Max results for cross-entity search |
contextManager.search.snippetTokens | 16 | Snippet preview size in tokens |
Explanations
| Setting | Default | Description |
contextManager.explanation.expandContext | true | Expand surrounding code when explaining |
Orchestration
| Setting | Default | Description |
contextManager.orchestrator.enabled | true | Enable agent orchestration (registry + psmux send). Allows multiple CLI/VS Code/Claude sessions to coordinate. |
Context
| Setting | Default | Description |
contextManager.context.stubLines | 5 | Lines of code captured per anchor stub for knowledge cards |
Save As Card
| Setting | Default | Description |
contextManager.saveAsCard.smartMerge | true | Enable smart merge detection when saving knowledge cards |
| Setting | Default | Description |
contextManager.tools.backgroundMode | true | Run save/search/read tools silently without confirmation dialogs |
Experimental
| Setting | Default | Description |
contextManager.experimental.enableProposedApi | false | Enable proposed VS Code APIs for enhanced features like Smart Select embeddings. Requires VS Code Insiders. |