Issues

• Tracking: replica-safe defaults in one change (count cache, config cache, MCP, config API, chart)

  horizontal scalingkeep-open
  Task

  MemMachine/MemMachine#1758 · Filter by author edwinyyyu opened on Oct 2, 2026

• Tracking: declare a concurrency scope per component so single-process components stay usable without blocking multi-replica deployments

  horizontal scalingkeep-open
  Task

  MemMachine/MemMachine#1757 · Filter by author edwinyyyu opened on Oct 2, 2026

• Tracking: background work across replicas with one claimable-job primitive (ingestion, deletion, purge)

  concurrencyhorizontal scalingkeep-openrecovery
  Task

  MemMachine/MemMachine#1756 · Filter by author edwinyyyu opened on Oct 2, 2026

• Tracking: make the session lifecycle safe across replicas with a registry row and durable jobs

  concurrencyhorizontal scalingkeep-openpriority: highrecovery
  Task

  MemMachine/MemMachine#1755 · Filter by author edwinyyyu opened on Oct 2, 2026

• Deployment artifacts assume one replica: Helm hard-codes replicas: 1 with no affinity or worker count, compose cannot scale, and all replicas would share one log file

  documentationhorizontal scaling
  Task

  MemMachine/MemMachine#1754 · Filter by author edwinyyyu opened on Oct 2, 2026

• Declarative memory and the Neo4j semantic backend keep per-process caches that are not refreshed when another replica writes

  horizontal scaling
  Bug

  MemMachine/MemMachine#1753 · Filter by author edwinyyyu opened on Oct 2, 2026

• The config API mutates one process and rewrites the whole configuration file with secrets in plaintext

  horizontal scalingsecurity
  Bug

  MemMachine/MemMachine#1752 · Filter by author edwinyyyu opened on Oct 2, 2026

• MCP over HTTP is stateful: a session id issued by one replica is unknown to the others

  horizontal scaling
  Bug

  MemMachine/MemMachine#1751 · Filter by author edwinyyyu opened on Oct 2, 2026

• MCP default identity derives from the host MAC address, so a client that sends no ids gets a different project on every machine

  horizontal scaling
  Bug

  MemMachine/MemMachine#1750 · Filter by author edwinyyyu opened on Oct 2, 2026

• Session deletion runs in a per-process queue: a crash loses the job until a restart, and every replica re-enqueues every Deleted session at boot

  horizontal scalingrecovery
  Bug

  MemMachine/MemMachine#1749 · Filter by author edwinyyyu opened on Oct 2, 2026
  

• Short-term memory is process-local: recent episodes are never reloaded, replicas diverge, and the persisted summary is last-writer-wins

  concurrencyhorizontal scaling
  Bug

  MemMachine/MemMachine#1748 · Filter by author edwinyyyu opened on Oct 2, 2026

• Semantic config cache is per process and on by default: stale categories across replicas, and a stale replica overwrites a set's embedder and language model

  concurrencyhorizontal scaling
  Bug

  MemMachine/MemMachine#1747 · Filter by author edwinyyyu opened on Oct 2, 2026

• Episode count cache is per process and on by default, so episode counts go stale across replicas

  horizontal scaling
  Bug

  MemMachine/MemMachine#1746 · Filter by author edwinyyyu opened on Oct 2, 2026

• Semantic background ingestion runs in every replica with no claim, so N replicas ingest the same messages N times

  horizontal scaling
  Bug

  MemMachine/MemMachine#1745 · Filter by author edwinyyyu opened on Oct 2, 2026

• [Bug]: Milvus 2.6.8 through 2.6.25 and 3.0.0 drop rows for a property filter that ORs two ranges with equal lower bounds (milvus-io/milvus#52852)

  keep-openupstream
  Bug

  MemMachine/MemMachine#1744 · Filter by author edwinyyyu opened on Oct 2, 2026

• [Bug]: Qdrant over gRPC delivers a still-compressed reply about once per 70,000 calls, failing the call (grpc/grpc#42823, #43532)

  keep-openupstream
  Bug

  MemMachine/MemMachine#1743 · Filter by author edwinyyyu opened on Oct 2, 2026

• [Bug]: Qdrant 1.15.0 through 1.19.1 answers a tenant-plus-property filtered query with nothing or the wrong page (qdrant/qdrant#10903 through #10907)

  keep-openupstream
  Bug

  MemMachine/MemMachine#1742 · Filter by author edwinyyyu opened on Oct 2, 2026

• Episode add is not atomic across the store, episodic memory, and semantic history; partial failures are unrecoverable without a full scan

  recovery
  Task

  MemMachine/MemMachine#1738 · Filter by author edwinyyyu opened on Oct 1, 2026

• [Bug]: Deleting a project deletes the org-level semantic set every project shares, and keeps the project's tagged sets

  multiuser
  Bug

  MemMachine/MemMachine#1732 · Filter by author edwinyyyu opened on Oct 1, 2026

• [Bug]: Semantic delete_history filters on history_id, which no index covers, so every episode deletion scans all tenants' citations and history

  Bug

  MemMachine/MemMachine#1731 · Filter by author edwinyyyu opened on Oct 1, 2026

• [Bug]: Semantic session manager and semantic service each build their own config cache, so set type deletes and registrations do not reach ingestion

  Bug

  MemMachine/MemMachine#1730 · Filter by author edwinyyyu opened on Oct 1, 2026

• [Bug]: vector_store semantic search filters by set after a server-wide top 10,000, so a set's own features drop out of its search

  Bug

  MemMachine/MemMachine#1729 · Filter by author edwinyyyu opened on Oct 1, 2026

• [Bug]: Deleting a session whose memory is not open creates its storage first, only to delete it

  performancesimplification
  Bug

  MemMachine/MemMachine#1728 · Filter by author edwinyyyu opened on Oct 1, 2026

• Refuse a SQLite runtime older than 3.35 in every store that uses RETURNING

  Task

  MemMachine/MemMachine#1726 · Filter by author edwinyyyu opened on Oct 1, 2026

• pgvector semantic storage tests compare a host-clock cutoff with database-stamped times 10 ms apart

  Bug

  MemMachine/MemMachine#1724 · Filter by author edwinyyyu opened on Sep 30, 2026