Notion Cortex: A Multi-Agent AI Research System Using Notion as the Operating System
Notion Cortex is a multi-agent AI research system that uses Notion as its operating system, allowing agents to think, communicate, and await human approval within the Notion platform.
Why it matters
Notion Cortex demonstrates how Notion can be used as an operating system for multi-agent AI research systems, enabling real-time collaboration and human-in-the-loop workflows.
Key Points
- 1Notion Cortex uses 5 specialized AI agents (Scouts, Analyst, Synthesizer, Approval Gate, Writer) to research a given topic in parallel and produce a structured intelligence brief
- 2The agents' reasoning streams into Notion's Working Memory pages in real-time, allowing users to observe their thought process
- 3Key technical decisions include serialized knowledge graph upsert, a two-queue concurrency design, idempotent workspace bootstrapping, and a dual-provider LLM abstraction
Details
Notion Cortex is a multi-agent AI research system that leverages Notion as its operating system. It decomposes a given research topic into 5 parallel angles, with specialized agents (Scouts, Analyst, Synthesizer, Approval Gate, Writer) working together to produce a structured intelligence brief. The agents' reasoning streams into Notion's Working Memory pages in real-time, allowing users to observe their thought process. Key technical decisions include a serialized knowledge graph upsert to handle parallel entity discovery, a two-queue concurrency design to manage Notion API rate limiting and logical atomicity, idempotent workspace bootstrapping to handle schema migrations, and a dual-provider LLM abstraction to support both OpenAI and Anthropic. The system is built with 55 tests across 13 files to ensure comprehensive coverage of the orchestrator pipeline, agents, and Notion data layer.
No comments yet
Be the first to comment