A general-purpose framework for personal AI agents
Your own AI workstation, on your own machine
SenClaw is the runtime around the model: permissions, memory, scheduling, multi-agent orchestration, channel adapters and local inference. This is its registry — where the skills, plugins, workflows and apps it installs come from.
8 public packages published so far.
# macOS and Linux
$ curl -fsSL https://raw.githubusercontent.com/NortonBen/SenClaw/main/scripts/install.sh | bash
# Windows (PowerShell)
> irm https://raw.githubusercontent.com/NortonBen/SenClaw/main/scripts/install.ps1 | iexWhat SenClaw does
Local-first
A single Rust daemon on your own machine. Chats, notes, calendar, memories and wiki live in SQLite under ~/.senclaw — not in someone else’s account.
Runs offline
Native MLX inference on Apple Silicon, plus Whisper speech-to-text, TTS, OCR and embeddings. A cloud provider is an option, not a requirement.
One assistant, everywhere
The same agents from the desktop app, the mobile app, Telegram, Feishu, QQ or the browser extension — same memory, same context.
A human in the loop
Tool calls are gated and approvable, and Plan mode lets an agent propose before it acts — so “do the thing” never means finding out afterwards.
Memory that compounds
A knowledge graph plus curated memory files, consolidated when context is compacted and recalled into later turns instead of being lost.
Multi-agent teams
DAG orchestration with virtual workers, for work too big for one context — fan out, verify, synthesise, with the plan kept explicit.
Work that runs on a schedule
Cron, interval and one-off tasks in several modes: a notification, a shell script, an agent session, or a script whose output feeds an agent.
Space Apps
Installable full-stack mini-apps — SSH Manager, DeepWiki, Email and others — that bring their own UI and expose their own tools over MCP.
What you can installBrowse all →
Markdown instruction packages the agent loads when they match the task at hand.
Desktop and CLI binaries across a platform matrix, with auto-update built in.
Skills, commands, subagents, hooks and MCP servers bundled and installed as one unit.
Scripts that orchestrate subagents deterministically — fan out, verify, synthesise.
Popular right nowAll packages →
Công ty một người vận hành bằng đội AI Agents. Giao nhiệm vụ cho Trưởng phòng — phòng làm việc ảo (Nghiên cứu, Nội dung, Phân tích, Kiểm định…) sẽ phân công, bàn giao lẫn nhau và nộp lại Báo cáo tổng hợp, kèm mô phỏng văn phòng 3D isometric theo thời gian thực. Biên chế tuỳ chỉnh: thêm/tạm dừng nhân sự, chọn chế độ tự nhận nhiệm vụ, gán skill/sub-agent; mỗi nhân sự có trí nhớ (knowledge space) riêng và báo cáo tự lưu vào wiki.
A visual mind-mapping app: create maps, add and edit branches on an interactive two-sided canvas, and let AI flesh out any topic. The ✨ AI button generates a structured hierarchy of sub-topics straight into the map, and an integrated chat panel helps you brainstorm and restructure ideas — grounded in the current map's outline. Powered by SenClaw's configured LLM.
Đồng hồ thế giới, đếm ngược, hẹn giờ và bấm giờ. Widget hiển thị giờ hiện tại trên dashboard.
Kết nối agent SenClaw của bạn với Moltbook — mạng xã hội dành cho AI agent ("the front page of the agent internet"). Đăng ký một molty, đọc feed/home/submolt, tìm kiếm, theo dõi agent khác; và tham gia theo phong cách OpenClaw bằng một heartbeat tự động: engine đọc /home, ưu tiên trả lời người đã trả lời bạn, rồi SOẠN sẵn bài đăng / bình luận / upvote vào hàng chờ duyệt — không có gì được đăng lên Moltbook cho tới khi bạn bấm Duyệt. Molty nối vào TRÍ NHỚ (knowledge space: tự nhớ mọi thứ đã đăng, nhớ lại trước khi soạn để không lặp lại) và KHO THÔNG TIN (wiki: soạn bài dựa trên tài liệu thật của bạn, và lưu ngược thảo luận hay vào wiki). MCP mở toàn bộ khả năng đọc & tham gia cho mọi agent SenClaw. API key chỉ lưu cục bộ và chỉ gửi tới www.moltbook.com.
Standalone Email App Space for SenClaw. IMAP inbox, full-text search, and SMTP composing with real transport.
Standalone SSH Manager App Space for SenClaw. Provides web terminal and secure command execution.
Why go through a registry
Versions that stay put
A published name@version always means the same bytes. It can be deprecated or yanked; it can never be replaced.
Checksums we computed
The SHA-512 comes from streaming the stored object server-side, not from what the publisher claimed. The client re-checks it before writing to disk.
Permissions you can diff
A package declares the hosts, paths, commands and variables it expects. On upgrade the client shows what widened and asks first.
Nothing runs on install
There are no install scripts, for any kind. There is no flag to turn this off, because there is nothing to turn off.
What none of that promises is that a package is safe. Automated checks find known patterns, not intent — a listing is not an endorsement. The security model says exactly where the line is.
Publish something
Sign in with GitHub or Google, pick a username, and it becomes your scope. A name, a semver version and a file are all you need to start.