PROJECT · INTERNAL · LIVE IN PRODUCTION

Meridian

A personal AI operating system. Live in production. Built and governed solo across 47 sessions.

THE NUMBERS
144+ decisions locked, each with rationale and rejected alternative documented
47 sessions handled with zero context loss between them via the deep-context protocol
9 canon principles — cross-project governing law applied to every engagement
0 fabrication events since the governance architecture was implemented

Meridian is a personal knowledge system — email corpus, knowledge graph, AI extraction pipeline. The data is private. The governance architecture is public. The patterns that emerged from 47 sessions of building a production AI system from scratch are the artifact.

Nine principles. Each one applies cross-project, to every engagement, without exception.

  1. Fabrication is not mitigated by correcting it afterward — it is prevented by architecture.
  2. Extraction is proposal, not fact — human authorization required before production state.
  3. Every runtime class holds exactly one credential with minimum required scope.
  4. Tool descriptions are compiled behavioral artifacts, not documentation.
  5. Decisions are locked with rationale and rejected alternatives — never quietly reversed.
  6. Pre-flight before execution — read-only analysis gates every code-changing prompt.
  7. Governance before code — the legal and architectural questions precede the sprint.
  8. A system you can trust must be a system where you can surgically revoke access.
  9. Model selection is a behavioral contract evaluation, not a benchmark comparison.
MIT LICENSED · charleskjohnson.com