gbrain vs Supermemory
Two of the top projects in Memory, side by side: who leads on each score part and who each one suits, then setup, license, activity and what each review found.
gbrain
Memory store with sources and a knowledge graph for coding agents
Supermemory
Memory and context engine for AI agents, with user profiles and hybrid search
Verdict
gbrain ranks higher, 91 to 89. Of the 4 score parts we measure for both, gbrain leads on 2, Supermemory on 1 and 1 is even.
- Adoptiongbrain leads by 2 points, 81 to 79: close to even.
- FreshnessEven at 100.
- MaintenanceSupermemory leads by 1 point, 94 to 93: close to even.
- Agent-readygbrain leads by 15 points, 85 to 70: a wide gap.
Pick gbrain if you are
Engineers giving Claude Code, Codex or other agents persistent, sourced memory
- Keyless start: embedded PGLite and keyword search, no server or API keys needed
- MCP access via local stdio, Tailscale tailnet exposure, or remote URL with token
Pick Supermemory if you are
Developers adding long-term memory to agents, apps and AI coding tools
- Hybrid search returns RAG documents and user memories in a single query
- Handles knowledge updates and contradictions, and auto-forgets expired facts
| What we compare | gbrain | Supermemory |
|---|---|---|
| Score parts, out of 100 | ||
| Adoption | 81, widely used | 79, popular |
| Freshness | 100, active | 100, active |
| Maintenance | 93, healthy | 94, healthy |
| Agent-ready | 85, ready | 70, partly |
| Facts from GitHub and the README | ||
| Stars | 32.7k | 31.6k |
| License | MIT (permissive) | MIT (permissive) |
| Last commit | Nov 2026 | Nov 2026 |
| Last release | Nov 2026 | Oct 2026 |
| Language | TypeScript | TypeScript |
| Docker | Yes | No |
| GPU | Not needed | Optional |
| arm64 or Apple Silicon | Mentioned | Mentioned |
gbrain
GBrain stores facts and markdown pages with their sources and exposes them over MCP to agents such as Claude Code, Codex, Cursor and ChatGPT. It starts keyless with keyword search on embedded PGLite, and optionally adds embeddings, reranking, a typed link graph, and `gbrain think`, which writes cited answers with gap notes. Postgres and a paid hosted option at gbrain.io are also available.
Who it is for: Engineers giving Claude Code, Codex or other agents persistent, sourced memory
Strengths
- Keyless start: embedded PGLite and keyword search, no server or API keys needed
- MCP access via local stdio, Tailscale tailnet exposure, or remote URL with token
- `gbrain think` answers carry citations and note stale or missing knowledge
- Swappable schema packs and a typed link graph; MIT license
Weaknesses
- Not on npm; install from GitHub with Bun 1.4.4 or newer
- Semantic search, synthesis and enrichment need separately configured, paid model providers
- Shared brains are not isolated against local file or shared DB credential access
- Always-on enrichment suggests a server with 8GB+ RAM plus its own API usage
- RAM ≥ 8 GB
- no GPU
- Compose
- Needs Bun 1.4.4+, PGLite or Postgres, Tailscale (optional, remote MCP), Embedding and chat provider API (optional)
Supermemory
Supermemory extracts facts from conversations, tracks updates and contradictions, expires stale information, and returns a per-user profile plus relevant memories through one API. It also covers RAG, connectors (Google Drive, Gmail, Notion, OneDrive, GitHub) and file extraction. It can be used via the hosted API, an MCP server and editor plugins, or a local server on port 6767.
Who it is for: Developers adding long-term memory to agents, apps and AI coding tools
Strengths
- Hybrid search returns RAG documents and user memories in a single query
- Handles knowledge updates and contradictions, and auto-forgets expired facts
- Local server keeps data in ~/.supermemory and can use Ollama with local embeddings
- Plugins for Claude Code, Cursor, Codex, OpenCode, plus Vercel AI SDK, LangChain, Mastra wrappers
Weaknesses
- Benchmark #1 claims are self-reported in the README, with details on a separate research page
- Hosted MCP server and plugins point at supermemory.ai cloud by default
- Local mode URL ingestion depends on a hosted reader service
- Local-versus-Enterprise feature differences are only documented on a separate page
- GPU optional
- Needs LLM provider (OpenAI, Anthropic, Gemini, Groq, or Ollama), Docker (optional)
- Models: OpenAI, Anthropic, Gemini, Groq, Ollama
- port 6767
More in Memory
Building with these?
We help teams pick an open-source AI stack and run it in production. Client work never changes a rank.