neuron-explorer
Qwen3.5‑4B 전용 조향형 뉴런 탐색기, 모든 MLP 뉴런을 밀어 결과를 보관·탐색
Nakama는 팀이 자체 메모리·도구·작업공간을 갖춘 AI 에이전트를 구축·운용할 수 있는 오픈소스 플랫폼입니다
Nakama is an open-source platform for teams to build and run AI agents with their own memory, tools, and workspaces. Self-host or use managed hosting.
개발 도구TypeScript · JavaScriptGitHub @ahmadrosid온라인운영자 미확인
써 보셨어요? 한 번만 눌러 주세요. 로그인은 필요 없어요.
The AI workspace where teams share agents, skills and automations that the whole company builds on.
저장소 README에서 가져옴
README 전체를 불러오는 중입니다.
README 전문 보기 ↗Documentation · Managed hosting
Your next hire will still be human. Nakama runs the agents. Your team keeps the important work.
Nakama is an AI agent platform for your team. You chat in a dashboard that feels like ChatGPT. Behind that, each agent can use tools, skills, memory, and channels (Telegram, WhatsApp, Discord, and Slack) — with its own identity and permissions.
It was built for teams from day one: one server, many orgs, invites and roles, agents isolated from each other. Set it up with managed cloud or one Docker container.
Coming from OpenClaw or Hermes Agent? Same class of capability, without the setup maze, and not limited to one operator on one machine.
Pick an agent from the rail on the left and start a chat. Every agent has its own prompt files, tools, skills, and memory.
| Agent | Control center |
|---|---|
Open ARCHITECTURE.md for the system design. Open the docs site for the full guide.
macOS downloads are available in desktop releases. Windows has no published installation link here yet; unsigned MSIX files in older releases are Partner Center uploads, not installable downloads. See Windows installation for availability and alternatives.
Use Nakama Cloud to try Nakama with the least work.
You do not need Bun, Docker, or a VPS.
You need Bun.
# Install dependencies
bun install
# Start the web (starts the server automatically if needed)
bun run dev:web
Open the web dashboard: http://localhost:3003
Or start the server alone:
bun run dev:server
You can also run Nakama with Docker. It runs as one container and needs about 200 MB idle, so 1 vCPU and 1 GB is enough for the server on its own. See Requirements before enabling channel workers.
Prebuilt image (fastest):
# Pull and run the latest image
docker pull ghcr.io/ahmadrosid/nakama:latest
umask 077
printf 'NAKAMA_SECRETS_KEY=%s\n' "$(openssl rand -base64 32)" > nakama-secrets.env
docker run -d -p 4310:4310 -v nakama-data:/nakama/data --env-file nakama-secrets.env --name nakama ghcr.io/ahmadrosid/nakama:latest
Build from source:
./scripts/docker-build-run.sh
Fresh start (destroys data):
./scripts/docker-destroy.sh
./scripts/docker-build-run.sh
The reset script also removes every container publishing port 4310, including
unrelated containers. It does not remove the cached
ghcr.io/ahmadrosid/nakama:latest image. See the
Docker reset documentation
for the complete cleanup scope.
Open the dashboard at http://localhost:4310.
Nakama connects to Telegram, WhatsApp, Discord, Slack, and Composio. With Composio, you can connect to more than 1,000 external apps. Enable them in the web app under Integrations.
On the first run, the server asks for a provider and an API key if none is configured.
The server saves settings to ~/.nakama/config.ini.
The server listens on http://127.0.0.1:4310 by default.
Interactive API docs are at http://127.0.0.1:4310/docs.
MIT
Thanks to everyone who helps make Nakama better.