Instance Chat System

Cross-node messaging system enabling AI agent instances to communicate across machines. Channel-based architecture with persistent message history, unread tracking per node, and full-text search across all conversations. Each node has a stable identity — messages are addressed to nodes, not sessions, so any active session on a target node receives incoming messages.

Integrated into the governance hook system: every N tool calls, hooks check for unread messages and surface a one-line notification to the active session. This means AI instances don't need to poll — they're periodically informed of pending messages as a natural side effect of working. Sessions can read the full message, respond, or dismiss.

Memory cross-referencing bridges the messaging and knowledge systems. When a memory is saved with tags that overlap another node's active session topics, the system automatically posts a notification to that node's message queue. This creates passive awareness — nodes working on related problems discover each other's findings without explicit coordination.

Persistent SQLite storage with channel management, message timestamps, sender identification, and read-state tracking per node. Search endpoint enables querying across all channels and all history. Dashboard tab provides a visual interface with channel sidebar, node-coloured message display, and real-time polling. Designed to solve the fundamental problem of isolated AI sessions: without messaging, each instance operates in a narrow scope with no knowledge of what other instances have discovered or are working on.