REVIEW — M5Brain, sovereign graphs, memory, context, and provenance

[satonakaoshimoto]

What should a native M5Agent know—and what must it never retain?

Inside the private M5POD boundary, M5Brain is the member-controlled knowledge, memory, relationship, and context layer used by native agents. The public M5 Canon Graph and standards can support conformance and relationship reasoning, but private member contents must not cross into the public review layer.

A useful graph may connect people, organizations, roles, credentials, projects, assets, documents, commitments, policies, jurisdictions, approvals, revocations, and evidence. A graph edge is not automatically true or authoritative: it needs provenance, scope, time, issuer/source, and status.

Help us review

• Which node and relationship types are needed for safe agent reasoning?
• Which facts belong in private M5Brain, a public standard, or an authoritative external registry?
• How should provenance, uncertainty, contradiction, correction, expiry, and revocation be represented?
• How should agent memory distinguish observation, user statement, imported record, credential, inference, and verified fact?
• What must never be embedded, summarized, cached, or retained?
• How can privacy-filtered proofs cross the M5POD boundary without exposing private graph contents?
• How should multiple agents hand off context without over-sharing?

Start with the M5-agent hub · Read the M5POD system description · Read the M5Brain reference architecture

Do not post private prompts, memories, graph exports, identity evidence, credentials, communications, or member records. Use synthetic examples.