Memory & tools

Memory & tools

The parts that make an agent feel like a colleague instead of a prompt: what it remembers, what it can run, and what happens when either one is wrong.

Vintage IBM 2401 magnetic tape drives with open reels
Photo: Adam Schuster· CC BY 2.0· Wikimedia Commons

Concept pages

  • Three ways to give an agent memory — coming
  • What is actually in the repository — coming
  • Five alternatives, and when to switch — coming
  • Hermes Agent vs OpenClaw — coming, after we run OpenClaw

Under the hood

  • Skills: the file format and what gets loaded when — coming
  • Scheduled tasks — coming
  • Approval gates and command allowlists — coming
  • Cutting toolsets to cut tokens — coming
  • Wiring an MCP server — coming
  • Running your own client against the API server — coming
  • Keeping memory in sync across machines — coming
  • Background processes and why a prompt never returns — coming