February 2026 Summaries

Supermemory announced an updated OpenClaw plugin designed to address perceived limitations in OpenClaw’s built-in memory system, which stores daily logs and curated long-term facts in Markdown files and relies on agents explicitly invoking search and retrieval tools. The post argues that this tool-based approach can add latency and token costs, retrieve context only when requested, create redundant records, and inadequately manage knowledge updates, temporal relationships, multi-session context, and outdated information. Supermemory instead uses background hooks to automatically capture both extracted memories and raw conversation chunks, supported by a vector-graph architecture intended to connect facts, update knowledge contextually, and decay irrelevant information over time. In benchmarks conducted using the company’s open-source Memory Bench platform, Supermemory reported an 85.9% score, compared with 58.3% for OpenClaw-style RAG memory and 54.2% for filesystem-based memory, while claiming lower token use and faster operation. The plugin is available through GitHub and is offered as part of a $20-per-month premium memory service.

Supermemory is a plugin for Claude Code and OpenCode designed to give coding agents persistent, scoped memory across sessions and projects. After a one-time `/supermemory-init` research process that examines a repository’s structure, history, conventions, and recurring issues, it stores project knowledge alongside cross-project user preferences and semantically searchable memories. It automatically injects relevant profile details, project context, and query-related memories at the start of each session, while allowing agents to save, search, browse, or delete memories through natural-language triggers or dedicated tools. The plugin distinguishes user and project memories using hashes derived from Git email and working directory, supports redaction of content enclosed in private tags, and uses configurable settings for retrieval and capture behavior. To avoid context-window degradation, it performs preemptive compaction at a configurable threshold, preserving requests, progress, remaining work, and explicit constraints in searchable project summaries that can restore work in later sessions.