Roadmap

The memory layer, in five phases.

rMem is a memory-layer platform. Today, MemoryPort handles import, search, and export. Each phase below adds a new capability to the same Postgres-backed memory store.

  1. Phase 0 — MemoryPort

    In progress

    Import, search, and export AI conversations across providers. Core flows shipped; billing (Stripe checkout, customer portal, plan reconciliation) still outstanding before Phase 0 closes.

    IssuesSWR-36SWR-23SWR-24SWR-70SWR-71
  2. Phase 1 — Semantic retrieval

    In progress

    pgvector embeddings, hybrid full-text + vector ranking, contextual chunk summaries.

    IssuesSWR-76SWR-77SWR-78SWR-79SWR-85SWR-88
  3. Phase 2 — Memory API + MCP

    Planned

    REST API and MCP server (stdio + HTTP) so any agent can read and write memory.

    IssuesSWR-80SWR-81SWR-82SWR-83SWR-84SWR-86
  4. Phase 3 — Structured memory

    Planned

    Entity extraction, dedup, memory management dashboard.

    IssuesSWR-87SWR-89SWR-93
  5. Phase 4 — Temporal & graph intelligence

    Planned

    Entity relationship graph, GraphRAG queries, temporal facts and decay.

    IssuesSWR-90SWR-91SWR-92

Issue identifiers reference internal Linear tickets. They are not deep-linked because the project is private.