Designs

Designs

The high-level design docs behind Agent M — its memory layer and the harness tooling around it — the full HLDs for what shipped and what's in flight, where the design started and where it's going. The Architecture pillar overviews link down into these for the deep detail.

MemoryVault

The permanent agent-memory store and its six implementation parts.

• MemoryVault — the design for durable, file-based agent memory.
  • write-primitives — how entries are captured and shaped.
  • recall-loop — how the right entries surface within a token budget.
  • reflection-and-recovery — self-maintenance + crash recovery.
  • idea-ledger — the incubator for not-yet-acted-on ideas.
  • seed-pass — the co-created first pass that bootstraps a vault.
  • discovery-mining — surfacing latent structure from existing notes.

Evolution & architecture

• Agent Memory Evolution V1→V4 — the full HLD: where Agent M started, how it grew, where it's going.
• Device-Wide Architecture — the device-wide design Agent M targets.
• Memory-OS Architecture (V5) — the V5 memory-OS HLD.
• The persona tier — the third classification above the substrate/plugin binary: a standing concern that composes capabilities it does not own (rememberer = degenerate persona; V5-11 chief-of-staff = first real persona). Refines ADR 0011 · pairs with ADR 0016.

Documentation & tooling

• Seven-Section Wiki Convergence — converging agentm's documentation spec + tooling onto the seven-section wiki taxonomy crickets standardized (amend ADR 0004 · retire the duplicate diataxis-author copy toward crickets · reshape templates/wiki/).

See also

Architecture · Decisions · Home