GitHubGitHub RepoNeeds Review

supermemoryai/supermemory

Supermemory combines persistent agent memory, profiles, RAG, connectors, and local deployment behind one API, with plugins for Codex, Claude Code, Cursor, and other clients.

GitHub

Open Source Open Markdown Open JSON

Source Summary

Supermemory combines memory extraction, maintained user profiles, hybrid RAG, connectors, and file processing behind one API, MCP server, and coding-agent plugins. It reports **95% Recall@15**, about **720 added tokens**, and **99.4% context reduction** on LongMemEval.

Practical Implication

Builders can use project-scoped containers to carry preferences and repo context between agent sessions, or run the same API locally at localhost:6767. The practical test is whether retrieved memories improve real tasks without stale facts, cross-project leakage, or prompt bloat.

Agent-Ready Context

Supermemory combines memory extraction, maintained user profiles, hybrid RAG, connectors, and file processing behind one API, MCP server, and coding-agent plugins. It reports **95% Recall@15**, about **720 added tokens**, and **99.4% context reduction** on LongMemEval.

Builders can use project-scoped containers to carry preferences and repo context between agent sessions, or run the same API locally at localhost:6767. The practical test is whether retrieved memories improve real tasks without stale facts, cross-project leakage, or prompt bloat.

The benchmark leadership and efficiency figures are project-reported, and the material does not describe independent validation. Automatic contradiction handling, expiry, and forgetting also need workload-specific evaluation before the memory layer becomes trusted state.

Connected Context · Feed7 Judgment

Supermemory packages several previously separate memory patterns into one deployable layer: extraction, maintained profiles, retrieval, connectors, and project-scoped continuity. Its reported efficiency strengthens the case for selective external memory, but does not settle the central trust problem in the prior candidates: persistent state still needs provenance, contradiction handling, expiry, isolation, and workload-specific evaluation before agents can rely on it.

thedotmack/claude-memBoth provide external cross-session project memory, but Supermemory broadens the pattern from inspectable activity history to profiles, connectors, file processing, and a hosted or local API.Lessons from Studying Every Memory System — Shlok Khemani, IndependentThe talk supplies the governance requirements that Supermemory’s maintained profiles leave unresolved, including conflict detection, visibility, editing, and deliberate update cadence.nashsu/llm_wikiLLM Wiki offers a more source-traceable form of persistent context, highlighting that Supermemory’s broader extraction and profile layer must still prevent durable summaries from preserving distortions.UniMem: Complementary Episodic-to-Parametric Memory for Boundary-Agnostic Task StreamsUniMem extends the design space beyond Supermemory’s retrieval-centered service by proposing learned consolidation for recurring patterns, while inheriting the same unresolved safety and inspectability concerns.

Context Map

agentcodingdata#agent-memory #context-engineering#retrieval

Uncertainty

The benchmark leadership and efficiency figures are project-reported, and the material does not describe independent validation. Automatic contradiction handling, expiry, and forgetting also need workload-specific evaluation before the memory layer becomes trusted state.