claude-mem vs gstack
Two of the top projects in Skills, plugins and prompts, side by side: who leads on each score part and who each one suits, then setup, license, activity and what each review found.
claude-mem
Persistent cross-session memory plugin for Claude Code and other coding agents
gstack
Markdown slash-command skills that give Claude Code planning, review, QA and ship roles
Verdict
claude-mem ranks higher, 97 to 96. Of the 4 score parts we measure for both, claude-mem leads on 1, gstack on 1 and 2 are even.
- AdoptionEven at 98.
- FreshnessEven at 100.
- Maintenancegstack leads by 1 point, 97 to 96: close to even.
- Agent-readyclaude-mem leads by 15 points, 85 to 70: a wide gap.
Pick claude-mem if you are
Claude Code users who want session memory that persists across runs
- Local SQLite plus Chroma storage with a worker HTTP API and web viewer
- Three-step search/timeline/get_observations flow filters before fetching full details, cutting tokens
Pick gstack if you are
Claude Code users wanting structured planning, review, QA and release workflows
- Claude Code is the only full-tier host; safety hooks like /careful and /freeze are enforced there
- Skills are plain Markdown, so they can be forked and edited
| What we compare | claude-mem | gstack |
|---|---|---|
| Score parts, out of 100 | ||
| Adoption | 98, widely used | 98, widely used |
| Momentum | Not scored for this kind | Not scored for this kind |
| Freshness | 100, active | 100, active |
| Maintenance | 96, healthy | 97, healthy |
| Agent-ready | 85, ready | 70, partly |
| Facts from GitHub and the README | ||
| Stars | 116.1k | 141.2k |
| License | Apache-2.0 (permissive) | MIT (permissive) |
| Last commit | Nov 2026 | Nov 2026 |
| Last release | Nov 2026 | Nov 2026 |
| Language | TypeScript | TypeScript |
| Docker | Yes | No |
| GPU | Not needed | Not needed |
| arm64 or Apple Silicon | Not stated | Mentioned |
claude-mem
Claude-mem hooks into Claude Code's lifecycle events (SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, Stop, SessionEnd) to capture tool-usage observations, summarize them, and inject relevant context into later sessions. Data is stored in SQLite with FTS5 and Chroma for hybrid search, served by a local Bun-managed worker with a web viewer. Search is exposed through three MCP tools (search, timeline, get_observations). Installers also exist for OpenCode, T3 Code, Antigravity, OMP, Pi and OpenClaw.
Who it is for: Claude Code users who want session memory that persists across runs
Strengths
- Local SQLite plus Chroma storage with a worker HTTP API and web viewer
- Three-step search/timeline/get_observations flow filters before fetching full details, cutting tokens
- Installers for several harnesses besides Claude Code, including OpenCode, T3 Code and OpenClaw
- `<private>` tags exclude sensitive content from storage
Weaknesses
- Installer prompts for sign-in to a hosted observer with a 30-day trial; opt-out needs flags
- Depends on Node 20+, Bun and uv, auto-installed if missing
- `npm install -g claude-mem` gives only the SDK, not the hooks or worker
- Some features, such as TeamBrain and cloud sync, rely on the hosted cmem.ai service
- no GPU
- Docker + Compose
- Needs Node.js >=20, Bun, uv, SQLite, Chroma
- Models: Anthropic plan, OpenRouter, Gemini, claude-mem observer
gstack
gstack is a set of Markdown skills, exposed as slash commands such as /office-hours, /plan-ceo-review, /review, /qa, /cso and /ship, that split a coding session into roles: product, architecture, design, review, QA, security and release. Setup is a git clone into ~/.claude/skills plus a Bun-based ./setup script that also builds a bundled browser for QA. Claude Code is the full-tier host; Codex, OpenCode, Cursor, Kiro and others are experimental.
Who it is for: Claude Code users wanting structured planning, review, QA and release workflows
Strengths
- Claude Code is the only full-tier host; safety hooks like /careful and /freeze are enforced there
- Skills are plain Markdown, so they can be forked and edited
- Team mode auto-updates for shared repos with no vendored files
- Installs for multiple agents (Codex, Cursor, OpenCode, Kiro, Copilot CLI) via --host flags
Weaknesses
- Non-Claude hosts are experimental or instruction-only, and safety skills only advise there
- Outside reviews need a second CLI (Codex or Claude Code) installed and signed in
- /cso needs a recent Bun with specific build flags plus a native C toolchain
- README leads with the author's productivity claims, not behavior of the skills
- no GPU
- Needs Claude Code, Git, Bun 1.3.3+ (1.4.2 tested), Node.js (Windows only), Codex CLI (optional, outside reviews), OpenAI API key (optional, design binary)
- Models: Claude (Opus 5.5), OpenAI GPT-6 Astra, OpenAI GPT-6.1 Sol, OpenAI gpt-5.5 and gpt-image-2 (design binary)
More in Skills, plugins and prompts
Building with these?
We help teams pick an open-source AI stack and run it in production. Client work never changes a rank.