Skip to content

Design: long-lived memory and state model #4

Description

@gidutz

Campaigns run for weeks. In-context memory is insufficient. We need durable state that an agent can read on a fresh invocation and reason about.

Memory categories:

  • Action log (what was done, by whom, when, with what rationale)
  • KPI snapshots (time series of performance metrics per entity)
  • Account-level heuristics (what's worked, what hasn't, learned norms)
  • Pending approvals and HITL state

Open questions:

  • Storage model: relational, document, vector, or mixed?
  • Retrieval: how does an agent know what to load on start?
  • Privacy and account isolation
  • Memory compaction / forgetting strategy

Deliverable: memory schema and retrieval interface.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions