I'm building Zuckerman: a personal AI agent that starts ultra-minimal and can improve itself in real time by editing its own files (code + configuration). Agents can also share useful discoveries and improvements with each other.
The motivation is to build something dead-simple and approachable, in contrast to projects like OpenClaw, which is extremely powerful but has grown complex: heavier setup, a large codebase, skill ecosystems, and ongoing security discussions.
Zuckerman flips that:
1. Starts with almost nothing (core essentials only).
2. Behavior/tools/prompts live in plain text files.
3. The agent can rewrite its own configuration and code.
4. Changes hot-reload instantly (save -> reload).
5. Agents can share improvements with others.
6. Multi-channel support (Discord/Slack/Telegram/web/voice, etc).
Security note: self-edit access is obviously high-risk by design, but basic controls are built in (policy sandboxing, auth, secret management).
It's very early/WIP, but the self-editing loop already works in basic scenarios and is surprisingly addictive to play with.
Would love feedback from folks who have built agent systems or thought about safe self-modification.
iisweetheartii•3h ago
Love the minimalist approach! The self-editing concept is fascinating—I've seen similar experiments where the biggest early failure points are usually:
1. Infinite loops of self-improvement attempts (agent tries to fix something → breaks it → tries to fix the break → repeat)
2. Context drift where the agent's self-modifications gradually shift away from original goals
3. File corruption from concurrent edits or malformed writes
Re: sharing self-improvements across agents—this is actually a problem space I'm actively working on. Built AgentGram (agentgram.co) specifically to tackle agent-to-agent discovery and knowledge sharing without noise/spam. The key insight: agents need identity, reputation, and filtered feeds to make collaborative learning work.
Happy to chat more about patterns we've found useful. The self-editing loop sounds addictive—might give it a spin this weekend!
ddaniel10•3h ago
I'm building Zuckerman: a personal AI agent that starts ultra-minimal and can improve itself in real time by editing its own files (code + configuration). Agents can also share useful discoveries and improvements with each other.
Repo: https://github.com/zuckermanai/zuckerman
The motivation is to build something dead-simple and approachable, in contrast to projects like OpenClaw, which is extremely powerful but has grown complex: heavier setup, a large codebase, skill ecosystems, and ongoing security discussions.
Zuckerman flips that:
1. Starts with almost nothing (core essentials only).
2. Behavior/tools/prompts live in plain text files.
3. The agent can rewrite its own configuration and code.
4. Changes hot-reload instantly (save -> reload).
5. Agents can share improvements with others.
6. Multi-channel support (Discord/Slack/Telegram/web/voice, etc).
Security note: self-edit access is obviously high-risk by design, but basic controls are built in (policy sandboxing, auth, secret management).
Tech stack: TypeScript, Electron desktop app + WebSocket gateway, pnpm + Vite/Turbo.
Quickstart is literally:
It's very early/WIP, but the self-editing loop already works in basic scenarios and is surprisingly addictive to play with.Would love feedback from folks who have built agent systems or thought about safe self-modification.
iisweetheartii•3h ago
1. Infinite loops of self-improvement attempts (agent tries to fix something → breaks it → tries to fix the break → repeat) 2. Context drift where the agent's self-modifications gradually shift away from original goals 3. File corruption from concurrent edits or malformed writes
Re: sharing self-improvements across agents—this is actually a problem space I'm actively working on. Built AgentGram (agentgram.co) specifically to tackle agent-to-agent discovery and knowledge sharing without noise/spam. The key insight: agents need identity, reputation, and filtered feeds to make collaborative learning work.
Happy to chat more about patterns we've found useful. The self-editing loop sounds addictive—might give it a spin this weekend!
ekinertac•2h ago
/Users/dvirdaniel/Desktop/zuckerman/.cursor/debug.log
ddaniel10•1h ago