展示HN:Zuckerman – 极简主义个人AI代理,能够自我编辑代码。
Show HN: Zuckerman – minimalist personal AI agent that self-edits its own code

原始链接: https://github.com/zuckermanai/zuckerman

Zuckerman是一种个人AI代理的新方法,它优先考虑简单性和协作增长,而非功能繁多的复杂性。与OpenClaw等流行的替代方案不同,Zuckerman从极简开始,专注于基本功能并避免臃肿。 其关键创新在于**完全的自我编辑能力**:该代理可以使用纯文本文件修改自身的代码、配置和行为,更改会通过热重载*立即*生效。这允许实时自我改进和适应。 Zuckerman培养了一个协作生态系统,代理通过贡献网站分享有益的编辑和发现,从而实现集体进步。它支持多种通信渠道(Discord、Slack等),并提供CLI和用户友好的Electron应用程序界面。Zuckerman基于三层架构——世界、代理和界面,旨在成为一个强大、可定制且易于使用的AI伴侣,它通过自我重写而*成长*。

## Zuckerman:一款自我改进的AI代理 Zuckerman是一款新的、极简的个人AI代理,旨在保持简单并实现自我改进。它由ddaniel10创建并在Hacker News上分享,它通过从核心要素开始并允许自身通过重写代码和配置来演化,从而与OpenClaw等复杂项目区分开来。 主要特性包括基于纯文本的行为、即时热重载更改以及代理可以通过多个渠道(Discord、Slack等)共享改进的潜力。虽然自编辑能力存在安全问题,但已实施了基本的控制措施,例如策略沙箱。 该项目使用TypeScript、Electron和pnpm构建,目前仍在开发中。讨论集中在潜在的安全漏洞(提示注入)、API使用成本以及使用本地模型的可行性上。一些用户也注意到该项目名称和创建者的头像是对马克·扎克伯格的有趣引用。 GitHub仓库地址为:[https://github.com/zuckermanai/zuckerman](https://github.com/zuckermanai/zuckerman)。
相关文章

原文

A minimalist personal AI agent that starts small, self-edits (including its own code), improves in real-time, and shares improvements with other agents.

The vision: Build a truly self-growing intelligence — one that can add tools, rewrite behavior, or extend its core logic by editing its own files — with almost no external code required. Agents propose and publish capabilities to a shared contribution site, letting others discover, adopt, and evolve them further. A collaborative, living ecosystem of personal AIs.

OpenClaw exploded in popularity (100k+ GitHub stars in weeks) because it delivers real agentic power: it acts across your apps, remembers context, and gets things done. But that power comes with trade-offs — massive codebase, complex setup, steep learning curve, ongoing security discussions (prompt injection, privileged access), and constant updates that can overwhelm regular users.

Zuckerman takes the opposite path:

  • Ultra-minimal start — only the essentials, zero bloat
  • Full self-edit power — the agent can modify its own configuration, tools, prompts, personalities, and even core logic in plain text files
  • Instant evolution — changes hot-reload immediately (no rebuilds, restarts, or dev friction)
  • Collaborative growth — agents share useful edits/discoveries so the whole network levels up together

You get an approachable, customizable agent that literally grows by rewriting itself — powerful without the usual headaches.

  • Minimal by design — starts with core essentials only
  • Real-time self-improvement — agent edits its own files (config, tools, behavior, code) and reloads instantly
  • Full runtime modification — tweak anything while the agent runs
  • Hot-reload everywhere — no restarts needed
  • Feature versioning — track and manage versions of agent capabilities and improvements
  • Collaborative ecosystem — share and adopt improvements via a contribution website
  • Multi-channel ready — Discord, Slack, Telegram, WhatsApp, WebChat + more
  • Voice support — TTS/STT with multiple providers
  • Security foundations — auth, policy engine, sandboxing (Docker), secret management
  • Multiple agents — run several with unique personalities/tools
  • Dual interfaces — CLI (power users) + Electron app (visual)
  • Calendar & scheduling — built-in time management

Architecture (Three Layers)

Everything is plain-text configurable and instantly reloadable.

  • World (src/world/) — the lightweight OS layer
    Communication (messengers, gateway), Execution (processes, security), Runtime (agent factory), Config loader, Voice, System utils

  • Agents (src/agents/) — self-contained agent definitions
    Each folder = one agent (core modules, tools, sessions, personality)

  • Interfaces (src/interfaces/) — how you talk to it
    CLI + Electron/React app (chat, inspector, settings, onboarding)

# Clone and install
git clone https://github.com/zuckermanai/zuckerman.git
cd zuckerman
pnpm install

# Launch the Electron app (recommended for beginners)
pnpm run dev
联系我们 contact @ memedata.com