# supermemoryai/supermemory

Source: [GitHub](https://github.com/supermemoryai/supermemory)  
Feed7 permalink: https://feed7.dev/p/supermemory-0larjzg  
Published: Unknown  
Trust: Needs Review (needs_review)

## Why Included

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

## 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 across 812 accumulated Signals:

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-mem](https://feed7.dev/p/claude-mem-1uba088) — Both 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, Independent](https://feed7.dev/p/lessons-from-studying-every-memory-system-shlok-khemani-independent-0m3gyxb) — The talk supplies the governance requirements that Supermemory’s maintained profiles leave unresolved, including conflict detection, visibility, editing, and deliberate update cadence.
- [nashsu/llm_wiki](https://feed7.dev/p/llm-wiki-1dqmknq) — LLM 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 Streams](https://feed7.dev/p/2607-26017v1-1opv1da) — UniMem 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

- Layer: agent
- Domains: coding, data
- Topics: 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.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.