Hey PH — Victor here, maker of Notal. The pain: agents execute, but the plan dies in the chat. By the next run the context is gone and you rebuild the prompt from scratch. I was rebuilding the same repo-health prompt every Monday. How it works: write the plan as a markdown notebook with `notal` blocks. Each block becomes an activity with a role, cycle, and dependencies. Notal exposes current activities through MCP. Your agent calls `activities_list_actionable`, runs one item, calls `activities_log`, and repeats until the queue is empty. When the cycle returns, the work comes back. Differentiation: Notion and Heptabase have AI buttons. Obsidian is open but you assemble your own agent stack. Notal is the only notebook that is itself an MCP server — any agent (Claude, Cursor, LangGraph, Goose) connects the same way they connect to GitHub, but the resource is your operational plan. If you try it: which of the 4 marketplace prompts (repo health, post metrics, billing review, restaurant opening) did you run, and did the agent interview build something usable?

Notal

Notes become living operations

Notes become living operations