MOSKVbot × CORTEX

A reference architecture for a bot that does not just remember. It carries a portable, auditable memory passport: decisions, preferences, research, tasks, agent state, and trust evidence.

Agéntica layer API guide Security

The stronger framing

MOSKVbot should not add “memory” as a feature. It should become the first visible product proving CORTEX as a trust substrate for agentic systems. The commercial hook is simple: every serious agent needs continuity, but continuity without provenance becomes liability.

Memory passport Verified decisions Tenant-scoped recall Research crystals Handoff state Ledger evidence

Runtime flow

1. Ingest

User message, tool result, file, event, or external signal enters MOSKVbot.

2. Recall

MOSKVbot queries CORTEX for tenant-scoped preferences, active ghosts, prior decisions, project context, and research facts.

3. Act

The model responds or triggers tools with explicit context, not hidden prompt-window folklore.

4. Guard

Candidate memories pass validation: schema, tenant, taint, secrets, confidence, and write-path policy.

5. Persist

Accepted facts are written to the CORTEX ledger with timestamps, provenance, source, and hash continuity.

6. Verify

Critical decisions can be exported, replayed, checked by consensus, or used in handoff to another agent/runtime.

Memory types

Not everything is “memory”. MOSKVbot should separate what it stores.

preferencedecisionknowledgeghosterrorhandoffresearchpolicy

Trust levels

Every remembered fact needs a status so the bot knows whether it can act on it.

observeduser-confirmedguardedconsensusverifiedexpireddisputed

Research mode

When MOSKVbot investigates, it should persist sources, claims, contradictions, and synthesized conclusions separately.

source → claim → synthesis → verified fact

Agéntica mode

When multiple bot agents collaborate, CORTEX becomes shared ground truth with heartbeat, votes, conflict handling, and handoff.

agent → vote → ledger → recall

Why this is better than just adding Mem0

Question	Memory API answer	CORTEX answer
What does the bot remember?	Relevant prior context.	Typed facts with provenance, tenant scope, confidence, and temporal validity.
Can we prove when it learned it?	Usually vendor/log dependent.	Ledger timestamp, hash continuity, and exportable evidence.
Can a bad memory be isolated?	Often update/delete semantics.	Deprecate, dispute, tombstone, or require consensus before acting.
Can several agents share truth?	Memory sharing depends on the app.	Agent registry, heartbeat, WBFT voting, handoff, and shared ledger.
Can research survive chat loss?	Only if app persists it well.	Research crystals: source, claim, synthesis, confidence, and verification state.

Product wedge

The market pitch should be: “MOSKVbot is the first bot with a verifiable memory passport.” It remembers like a personal AI, audits like infrastructure, and hands off state like an operating system for agents.