Supermemory changelog

Better memories on the self-hosted server #

The self-hosted server now turns your documents into richer, more reliable memories, and uploads spend less time queued before they become searchable.

Improvements

  • Search on the local server now understands time, so you can ask for memories from last week or a specific period.
  • The local profile endpoint now returns memories organized into buckets instead of a flat list.
  • Search results now rank the memories you saved directly above inferred ones, which stay available further down.
  • PDF and spreadsheet uploads are now handled more predictably.
  • Telemetry can be turned off completely by setting SUPERMEMORY_DISABLE_TELEMETRY=1.
  • Upgrade with supermemory-server upgrade, or read the full release notes on GitHub.

Pluggable embeddings for the self-hosted server #

Choose the embedding model your local server uses: the built-in local model, OpenAI, any OpenAI-compatible endpoint, or Google.

Improvements

  • Pick a provider at first boot or set it with SUPERMEMORY_EMBEDDING_* environment variables; the server locks the choice so all vectors stay consistent.
  • Existing installs upgrade safely and keep their original local embedding model.
  • Switching to a different model with the same dimensions stops with a clear error instead of mixing incompatible vectors.
  • Upgrade with supermemory-server upgrade, or read the full release notes on GitHub.

Faster memory search during embedding provider slowdowns #

Memory search now avoids long retry chains, so queries stay under 600 ms even when an upstream embedding provider slows down.