Claude-Supermemory — Summary

Claude-Supermemory is a Claude Code plugin that delegates all memory storage and retrieval to the Supermemory cloud API (supermemory.ai). It ships 2 hooks (SessionStart for context injection, Stop for session save), 2 skills (super-search, super-save), and 3 slash commands (index, project-config, logout). The entire memory backend is remote: there is no local database, no local embeddings, and no local binary — just Node.js scripts that call the Supermemory API.

The cloud platform provides two memory containers per installation: personal container (per user, cross-session) and team container (per repo, shared across team members). Signal-keyword extraction controls what gets saved (remember, architecture, decision, bug, fix are the defaults). A maxProfileItems: 5 default limits how many memories are injected per session to avoid context bloat.

Compared to seeds, this is closest to ccmemory in concept (hook-driven session memory for Claude Code) but entirely cloud-dependent where ccmemory is self-hosted. The signal-keyword extraction approach (save only conversation turns containing specific keywords) differs from CSR (saves everything) and ccmemory (LLM detection of specific node types). Requires Supermemory Pro subscription ($) — the only paid-only framework in this batch alongside lean-ctx's cloud tier.

Claude-Supermemory — Overview

Origin

Supermemory.ai — the company behind the supermemory.ai platform. MIT license (though no license file found in repository; metadata says null). Version 0.0.4. JavaScript codebase.

Philosophy

From README:

"Enable Claude Code to learn in real-time, update it's knowledge, and grow with you, using supermemory." "Your agent remembers what you worked on - across sessions, across projects."

Core tenets:

• Team Memory — Project knowledge shared across team members, separate from personal memories
• Auto Capture — Conversations saved when session ends
• Signal Extraction — Only capture important turns (configurable via signalKeywords)

Design Choice: Cloud-Only

Unlike ccmemory (self-hosted Neo4j) or CSR (local binary), Claude-Supermemory explicitly delegates all storage to the Supermemory cloud API. This means:

• Zero local infrastructure
• Cross-device sync built-in
• Requires SUPERMEMORY_CC_API_KEY environment variable
• Requires Supermemory Pro subscription ($)

Key Configuration (settings.json)

{
  "maxProfileItems": 5,
  "signalExtraction": true,
  "signalKeywords": ["remember", "architecture", "decision", "bug", "fix"],
  "signalTurnsBefore": 3,
  "includeTools": ["Edit", "Write"]
}

The signalExtraction + signalKeywords pattern is the core filtering philosophy: instead of saving everything (like CSR) or using LLM detection (like ccmemory), Supermemory uses keyword matching to decide what gets saved. signalTurnsBefore: 3 captures the 3 turns of context before each signal keyword.

Claude-Supermemory — Architecture

Distribution

Claude Code plugin: /plugin marketplace add supermemoryai/claude-supermemory

Install Methods

/plugin marketplace add supermemoryai/claude-supermemory
/plugin install claude-supermemory
export SUPERMEMORY_CC_API_KEY="sm_..."

Directory Structure

plugin/
├── .claude-plugin/
│   └── plugin.json           # Plugin manifest
├── commands/                 # Slash commands
├── hooks/
│   └── hooks.json            # SessionStart + Stop hooks
├── scripts/
│   ├── context-hook.cjs      # SessionStart script (fetch memories)
│   ├── summary-hook.cjs      # Stop script (save session)
│   └── search-memory.cjs     # Search script (used by super-search skill)
└── skills/
    ├── super-search/SKILL.md
    ├── super-save/SKILL.md
    ├── supermemory-search/   # alias
    └── supermemory-save/     # alias

Config Files

• ~/.supermemory-claude/settings.json — global settings (maxProfileItems, signalKeywords, etc.)
• .claude/.supermemory-claude/config.json — per-repo overrides (apiKey, repoContainerTag)

Required Runtime

• Node.js (for .cjs scripts)
• SUPERMEMORY_CC_API_KEY environment variable
• Supermemory Pro subscription (https://app.supermemory.ai)

Memory Architecture

All memory lives in the Supermemory cloud:

• Personal container — per-user memories across all sessions
• Team container — per-repo memories shared across team (identified by repoContainerTag)

No local storage of memories. The plugin only stores config files locally.

Target AI Tools

Claude Code only (hooks and skills are Claude Code-specific).

Claude-Supermemory — Components

Hooks (2, from hooks.json)

Skills (4, with 2 aliases)

Skills call search-memory.cjs and similar scripts with a scope flag (--user, --repo, --both).

Slash Commands (3)

Scripts (3 in plugin/scripts/)

• context-hook.cjs — SessionStart: fetch top-K memories, format for injection
• summary-hook.cjs — Stop: extract and save session summary
• search-memory.cjs — Used by super-search skill; supports --user, --repo, --both scope flags

MCP Servers

None — all API calls go directly to Supermemory REST API from the Node.js scripts.

Subagents, Templates

None.

Claude-Supermemory — Uniqueness

Differs From Seeds

Closest seeds: ccmemory and claude-self-reflect — all three are hook-driven session memory layers for Claude Code. The critical delta: Supermemory externalizes the entire memory backend to a cloud API (supermemory.ai), while ccmemory uses self-hosted Neo4j + Docker and CSR uses a local Rust binary. This means Supermemory requires no local infrastructure and provides zero-config cross-device sync, but introduces a paid cloud dependency. Signal-keyword filtering (save only turns containing specific words) is a unique capture strategy vs ccmemory's LLM-based detection or CSR's capture-everything approach. The team memory container (shared per-repo across team members) is also distinct: ccmemory has a "team mode" for curated decisions but is single-server; Supermemory's team memory is built into the cloud platform.

Positioning

"The memory plugin for teams using Claude Code together — shared project knowledge that persists across developers, sessions, and devices, without running any infrastructure."

Observable Failure Modes

1. Cloud dependency — any Supermemory API outage or account termination destroys all memories with no local backup.
2. Keyword filter misses — signal-keyword extraction relies on users or AI explicitly using words like "remember" or "decision"; architectural context discussed without these keywords won't be captured.
3. 5-item injection limit — maxProfileItems: 5 may miss important context for large or long-running projects; but increasing it risks context bloat.
4. No API without subscription — requires Supermemory Pro; free users cannot use this plugin at all.
5. Privacy — all project context and conversations go to a third-party cloud API; not suitable for confidential/regulated codebases.