Chat Sessions
One row per session day or major pivot. Tracks active thread, open items, and session metadata.
rawshn@local : ~/hustle/kai-context-hub $
kai context hub
A continuity layer for multi-turn agent conversations: Notion as source of truth, local pointer for fast bootstrap.
- stack: Cursor + Notion MCP
- pattern: write-back every turn
- repo: private
What it is
Kai Context Hub is a memory and bootstrap system for long-running Cursor agent sessions. Every turn gets logged chronologically to Notion, with a local session pointer so a fresh chat can hydrate context instantly and resume from the last sequence number, without re-uploading exports or cold-starting from scratch.
Infrastructure only: no public dashboard, no shared workspace, no repo link. The design problem is agent continuity across weeks of multi-turn work in the same project.
Architecture
Three Notion databases plus a local fast cache.
-
-
Chat Log
Append-only chronological record. Each turn writes two rows: user message and agent response, keyed by sequence.
-
Memory Bank
Structured facts and follow-ups worth surfacing later. Priority-scored items hydrate on bootstrap.
Bootstrap flow
On session start, the agent reads local bootstrap files (config IDs, session pointer, system prompt), fetches the active session and recent messages from Notion via MCP, reconciles sequence numbers against the pointer, and responds with continuity. Write-back runs every turn so the local cache and Notion stay aligned.
Historical exports sit in cold storage and are not re-ingested on every bootstrap. The goal is zero cold-start friction for agent sessions that span many days.