Jarvis AI Assistant

Jarvis is the built-in AI assistant that operates within CoreLayer. It is not a standalone chatbot — it is a coordinator that routes requests through the Model Gateway, discovers tools via the Tool Registry, and enforces safety through the Permission Guard.

How Jarvis Works

User Input (voice, text, shortcut)
  → Intent interpretation
  → Context assembly (conversation history, memory, tools)
  → Model selection via Model Gateway
  → Response generation with tool calling
  → Permission check before tool execution
  → Result delivery

Capabilities

• Natural language task management — create, query, and update tasks
• Reading list management — track articles with AI-powered recommendations
• Daily/weekly reviews — summarize activity with pattern recognition
• Tool calling — execute tools from native modules, MCP servers, skills, or REST adapters
• Voice interaction — full voice pipeline with wake word, ASR, TTS, and interruption

Memory

Jarvis maintains long-term memory across conversations. Memory is stored locally in the configured database (SQLite by default) and persists across sessions.

• Conversation context — current session history, kept in memory during active conversations and written to the database on session end
• User preferences — learned patterns (e.g., preferred models, frequent tools, response style) accumulated over time
• Task context — active tasks, deadlines, and status — Jarvis references this when you ask about your work
• Tool context — recently used tools and their results, so Jarvis can follow up on previous actions without re-querying

Memory is scoped per user and never shared across CoreLayer instances unless you enable Supabase cloud sync.

Multi-Model Routing

Jarvis doesn't depend on a single model. The Model Gateway selects the best model for each request based on complexity, latency requirements, and availability.

Next Steps

• Control Center — the UI that hosts Jarvis
• Model Gateway — how models are selected
• Voice Pipeline — voice interaction setup