Functional memory
The short-lived working surface for the current session.
- Active task and immediate constraints.
- Fresh commitments and clarifications.
- Usually consolidated or cleared after the session.
Hypomnema is the scoped notebook of the relationship: durable enough to survive sessions, humble enough to revise, and explicit enough to promote only when it has earned a place in Mnemos.
Mnemos remembers by changing the graph. Hypomnema does something quieter first: it holds continuity in a scoped, revisable form. It is where an agent can keep track of what it is learning with a particular person or project without prematurely turning that material into a general memory, a belief, or a shared engram.
The short-lived working surface for the current session.
The relationship-scoped continuity layer.
The living graph and substrate.
A hypomnema entry moves through five states. The important thing is the pause: the system can carry continuity without pretending every note is already a settled memory.
Hypomnema is not a replacement for Mnemos. It is a buffer against overconfident memory formation, and a way to make continuity legible without freezing it too early.
Preferences, collaboration patterns, open threads, unresolved tensions, project-specific language, and continuity that matters only inside a particular relationship belong here first.
After a hypomnema entry becomes high-confidence, high-salience, foundational or revised, it can graduate into a de-identified engram.
Continuity changes. The system keeps revision history so the agent can remember how its understanding shifted, not just the latest wording.
A superseded entry remains in the audit trail but leaves active retrieval, which prevents stale context from speaking through the agent.
The new Mnemos MCP tools expose the layer directly. Agents can write continuity, search it, improve it, replace it, inspect promotion candidates, and explicitly graduate stable entries into Mnemos.
mnemos_hypomnema_write
Creates scoped entries with source, domain, tags, confidence, salience, density, and optional provenance.
mnemos_hypomnema_search
Returns entries by lexical relevance, confidence, salience, and foundational weight.
mnemos_hypomnema_revise
Updates content while preserving the prior version and the reason for change.
mnemos_hypomnema_supersede
Creates a successor entry and moves the original out of active retrieval.
mnemos_hypomnema_candidates
Lists active entries above confidence and salience thresholds that are revised or foundational.
mnemos_hypomnema_promote
Dry-runs by default, then writes a de-identified semantic engram when promotion is explicit.
The schema is deliberately boring. That is the point. Hypomnema should be inspectable, easy to query, and clear enough that a future agent can tell whether it is reading active continuity, stale continuity, or promoted continuity.
agent_id, person_id, project_scope
content, source, domain, tags
confidence, salience, density, foundational
active, superseded_by, graduated_to_engram_id
revision_count, revisions, last_revised_at
related_session_id, related_engram_id
Hypomnema lets Mnemos stay alive without forcing every continuity note to become a belief.