Skip to content

Vision Track: ELF as a high-trust memory system for single/multi-agent workflows #56

@aurexav

Description

@aurexav

Summary

Master tracking issue for ELF's next-stage direction as a high-trust memory system for single-agent and multi-agent workflows.

Direction Principles

  • Postgres remains source of truth.
  • Qdrant remains a derived, rebuildable retrieval index.
  • Evidence binding remains mandatory.
  • Memory quality is measured by replayable evaluation, not intuition.
  • Multi-agent collaboration uses explicit sharing, not implicit snooping.

Workstreams

Execution Order

  1. Graph Memory: canonical entity/fact temporal schema in Postgres #48
  2. Ingestion: evidence-bound entities/relations for add_note and add_event #49
  3. Knowledge correction: supersession and temporal validity for facts #52
  4. Search: graph relation context and explain integration #50
  5. Memory Policy v2: explicit remember/update/ignore/reject decisions #51
  6. Multi-agent memory spaces: explicit sharing and access grants #53
  7. Quality gate: CI regression checks with elf-eval trace compare #54
  8. Operator observability: memory inspector and diagnostics workflow #55

Stage Exit Criteria

Existing Related Work

Not a Commitment

This tracking issue describes direction and sequencing. Scope can be refined as implementation evidence emerges.

Metadata

Metadata

Assignees

No one assigned

    Labels

    kind:epicUmbrella issue that tracks multiple deliverables.theme:provenanceEvidence, citations, lineage, and explainability.theme:reliabilityCorrectness, consistency, failure handling, and operational robustness.

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions