server-v0.0.7

parruda/swarmserver-v0.0.7Aug 15, 2026by Dhravya

Upgrading?

AI Summary

A comprehensive release adding self-hosted licensing limits, a new local console interface, significant performance optimizations, and a diagnostic tool to check system health.

Key Highlights

  • Introduction of a 10,000-document cap for self-hosted licenses.
  • New local console featuring live counts, recent documents, and an interactive memory graph.
  • ~3s boot time achieved via lazy embedding model loading.
  • New 'supermemory-server doctor' diagnostic command.
  • Restored ONNX native runtime for ~12x faster embedding performance.

Breaking Changes

  • Multiple servers per machine - workflow engine state is now isolated per instance instead of shared globally.

New Features

  • 10,000 document cap enforcement
  • Local console with interactive memory graph
  • Lazy embedding model loading
  • Diagnostic tool (supermemory-server doctor)
  • ONNX native embeddings runtime
  • Automatic store healing for older stores
  • Multiple instances per machine support
  • Clean shutdown mechanism

Full Release Notes

The full 0.0.7 release — everything from the release candidates plus a hardening and polish pass across the board.

- **supermemory lite** — self-hosted is licensed up to 10,000 documents (read the license at git.new/memory); the cap is enforced at the API and shown at boot.
- **New local console** — the welcome page gets a Memory tab: live counts, recent documents and memories, container tags, and an interactive memory graph.
- **~3s boot** — the embedding model loads lazily in the background instead of blocking startup.
- **supermemory-server doctor** — one command that checks storage, keys, embeddings, the native runtime, port, and workflow engine, with concrete fixes for anything broken.
- **Native embeddings restored** — the ONNX native runtime ships inside the binary again (~12x faster than WASM, far lower memory).
- **Store healing** — upgrades from 0.0.5-era stores now migrate automatically instead of failing (#1325).
- **Multiple servers per machine** — each instance owns its workflow engine; no more shared global state on port 6420.
- **Clean shutdown** — Ctrl-C drains in-flight work, snapshots, and exits without panics.
- **Cheaper idle** — encrypted snapshots stream to disk and skip entirely when nothing changed.
- Plus fixes across API-key rate limiting, org reset, memory-generation failure handling, the embedding setup wizard, and Windows release builds.

Related Repositories