
JD
1.8K posts

JD
@jit_infinity
Started from nothing. I build for fun. Building → https://t.co/6Wk2BIWNvR






The biggest nanobot update ever just dropped 🚀 After two months of intense development, the community response has been incredible: - 1,866 commits pushing boundaries forward - Nearly 1,500 pull requests & 500+ issues resolved - 6.8k forks spreading across the ecosystem - 12 releases delivering continuous improvements - 158k downloads All powered by 220+ contributors who've driven the project to nearly GitHub 40k stars 🌟 GitHub: github.com/HKUDS/nanobot Huge thanks to Xubin @xubinrencs and the amazing community making this happen. We're not just building software – we're exploring the next-generation lightweight agent paradigm. These new features take nanobot from being a helpful AI assistant to becoming a dependable AI coworker that can tackle real challenges. Key Design of new nanobot: From AI Assistant to AI Coworker 1. 🔄 Long-Running Resilience → Handles multi-hour tasks without losing context or progress when interrupted. • Layered Context Management: Smart compression to avoid token window crashes • Large Output Handling: Disk persistence with reference replacement for efficiency • Smart Retry & Recovery: Classified error handling with checkpoint-based recovery 2. 🧠 Three-Layer Memory System → Learns your personal working preferences and evolves with your tasks over time, with full visibility into what it remembers. • Short-term Memory: Complete current session context with fast storage • Mid-term Memory: Auto-compression of conversations into structured summaries • Long-term Memory: Experience consolidation with git-tracked transparency 3. 🔒 Security Framework → Reduces common security risks when deploying AI agents in production environments. • Sandboxed Execution: bwrap containerization with non-root execution • Environment Isolation: No host variable leakage with localhost-bound ports • Configuration Security: ${VAR} interpolation eliminating hardcoded secrets 4. ⚙️ Extensible AI Kernel → Transforms nanobot from a standalone tool into a programmable AI kernel that integrates into various systems and workflows. • Embedded Design: 5-line integration with multi-session parallel support • Standard API Interface: One-line replacement for existing AI service integrations • Lifecycle Hooks: Developer observation and intervention at every execution step 5. 💬 Multi-Channel Communication → Works on platforms you already use with consistent functionality across channels. • Cross-Platform Coverage: Telegram, Slack, Discord, Feishu, WeChat, WhatsApp, Matrix, Email • Rich Media Processing: File uploads, voice transcription, video handling • Message Optimization: Auto-splitting, streaming output, intelligent replies 6. ⚡ Real-time Progress Streaming → Provides transparent progress updates during complex tasks for better user experience. • End-to-End Streaming: Complete streaming from providers through channels to CLI • Execution Visibility: Real-time display of agent thinking and execution process • Multi-step Task Tracking: Progress indication during long-running executions #nanobot #HKUDS

alright agent nerds, if you care about your tokens and usage limits, pay attention to the tools you give to your agents. i built a benchmark that compared various browser tools for agents, and here's an example of their massive difference in cost and latency doing the same task





My reddit marketing tool is now finding me opportunities, and clients 24/7 Made a new breakthrough today, now it works from anywhere without getting blocked by Reddit.















