mAIvn Docs

Platform

Portal Memory Management

Configure org memory policy and curate project skills, insights, and resources.

#Portal Memory Management

As your agents run, they accumulate memory: things they have learned, reusable procedures, and supporting documents. The Developer Portal is where you govern and curate all of that from a web UI: setting how much gets kept, cleaning up what is stale, and promoting the most valuable lessons so more of your agents can benefit.

This guide covers those memory operations in the mAIvn Developer Portal.

#Scope

Portal memory management is split across:

  • Organization-level policy and purge controls
  • Project-level memory assets (skills, insights, resources)

Use organization settings for hard governance.
Use project memory pages for operational curation.

Portal curation sits on top of the runtime memory split:

  • thread recall stays with the originating thread_id
  • skills, resources, and promoted insights are reusable assets scoped to agent, swarm, project, or org

#Organization Memory Policy

Open the Organization Settings page and scroll to the Memory Management section.

Controls:

  • memory enabled/disabled
  • persistence ceiling:
    • persist_none
    • vector_only
    • vector_plus_graph
  • optional retention windows:
    • vector retention days
    • graph retention days

These controls set ceilings enforced by the mAIvn service at runtime.
SDK memory_config can only request equal or lower behavior than the org ceiling.

#Organization Memory Purge

On the same Organization Settings page, scroll to the Memory Purge section.

You can purge persisted memory by:

  • organization scope (default)
  • project scope (project_id)
  • session scope (session_id)

Purge requires explicit confirmation token: PURGE_MEMORY.

#Project Memory Skills

Path: Project -> Memory -> Skills

Skills are reusable execution patterns with:

  • scope (agent, swarm, project, org)
  • structured steps
  • preconditions/postconditions
  • status lifecycle (active, deprecated, quarantined)

Use skills to stabilize repeated workflows and improve retrieval quality for known procedures.

#Project Memory Insights

Path: Project -> Memory -> Insights

Insights capture operational lessons and warnings with:

  • insight types (lesson, warning, optimization, failure_pattern)
  • relevance score
  • TTL / decay controls
  • staged scope promotion (agent/swarm -> project -> org)

Operational model:

  • auto-generated insights start narrow (agent or swarm)
  • promote to project when the lesson should benefit future threads across the project
  • promote to org only when the lesson is broadly reusable across teams

Tier note:

  • Insight management is available on Professional and Enterprise tiers.

#Project Memory Resources

Path: Project -> Memory -> Resources

Resource capabilities:

  • create from upload or source URL
  • update metadata/tags/scope
  • replace content (version chain)
  • bind to agent or swarm
  • soft-delete and restore
  • identify cleanup candidates (unbound/idle resources)

Binding types visible in the portal include portal, agent, swarm, message, and unbound.

Versioning is content-hash based. Re-uploading or binding identical bytes reuses the existing
non-deleted resource. Replacing content, or deploying an SDK-bound resource with the same
resource_id but different content_base64, creates a new registered version and marks the
previous active row superseded.

  1. Set organization memory ceiling first.
  2. Keep project defaults conservative (glimpse/focus) until needed.
  3. Curate high-value skills and resources by scope.
  4. Use insight promotion to move durable lessons from agent/swarm to project, then to org only when justified.
  5. Schedule periodic cleanup of unbound or superseded resources.