Skip to content

feat(observability): measure private-memory retrieval quality without memory content #120

Description

@robinbraemer

Outcome

Measure private-memory extraction, consolidation, retrieval quality, index health, and attachment amplification without exporting memory content.

Scope

  • Add bounded signals for extraction, Dream, retrieval method, candidates, selected topics/chunks, attachment bytes/tokens, latency, fallback, and degradation reason.
  • Record index schema/model freshness, rebuild outcomes, corruption recovery, and forget invalidation.
  • Link auxiliary provider calls to the owning Agent operation where possible.
  • Define offline evaluation metrics separately from production operational metrics.

Acceptance criteria

  • Operators can distinguish selector, lexical, vector, hybrid, and fallback paths.
  • Retrieval amplification and auxiliary model cost can be correlated to the main operation.
  • Topic paths, scopes, snippets, bodies, queries, embeddings, and model inputs are absent from exported telemetry.
  • Pause/forget telemetry proves invalidation occurred without revealing what was forgotten.
  • Memory instrumentation remains fail-open for the main turn.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions