Solve the problem first. The pitch can wait.

Working guides and one essay, each self-contained, most runnable on a laptop in a few minutes with nothing external. They use OMEM because that is what this site is, but every one solves its problem completely before it says anything a landing page would.

• EU AI Act Article 12 for AI agents: what to log, with working code

  The high-risk logging duty, read as engineering: why ordinary logs fail it, the four properties a compliant agent record needs, and a working self-hosted implementation.

• An audit trail for AI agents

  What a defensible trail must contain, why request logs do not qualify, and how to record what the agent believed, why, and who approved its actions.

• Human-in-the-loop approvals for AI agents

  Gate risky actions behind a named approver: a closed set of actions, risk decided by the system, and refusals kept on the record.

• Should an agent's memory decide what is true?

  The essay: overwriting on conflict is a truth judgment nobody chose, and the case for memory that keeps both sides and proves why.

• Long-term memory for a LangGraph agent

  Wire a LangGraph agent to a store where put supersedes instead of overwriting, delete retracts instead of erasing, and every memory can answer where it came from.

• A memory MCP server for Claude

  One command and one block of JSON give Claude Desktop remember, recall and why tools against a server running on your machine.

• Agent memory in Python, with receipts

  Beliefs instead of rows: contradictions surfaced, updates that keep history, retractions that take conclusions with them, and why() for everything.