Edster – An open-source local AI agent with swarm mode and a web UI
hackernews
|
|
📰 뉴스
#llama
#하드웨어/반도체
#ai
#review
#네이버웍스
#디지털 전환
#인공지능
#인하공업전문대
#직업교육
#평택시
#행정 혁신
원문 출처: hackernews · Genesis Park에서 요약 및 분석
요약
오픈AI가 기업용 AI 에이전트 SDK에 샌드박스 기능을 대폭 강화했다. 이번 업데이트는 에이전트를 특정 작업 공간 내에서 격리된 방식으로 작동시켜 필요한 파일과 코드에만 접근하게 함으로써 시스템 전체 무결성을 보호한다. 예측 불가능한 에이전트의 동작을 통제하기 위한 안전장치로, 프론티어 모델을 위한 인-디스트리뷰션 하네스도 제공한다. 새 기능은 파이썬으로 우선 출시되며 API를 통해 모든 고객에게 표준 가격으로 제공된다.
본문
An unstoppable, fully local, open-source coding agent that runs on your consumer GPU. Tags: ollama coding-agent local-ai cli rag chromadb python qwen Are you trying to use local LLMs to autonomously write code, read files, and manage your projects, only to watch them suffer from "tool amnesia," hallucinate XML tags, or get stuck in infinite execution loops? Nedster solves this instantly. Nedster is a highly autonomous, CLI-based AI software engineer designed for privacy-conscious developers. Powered by Ollama and augmented with TurboQuant 4-bit KV Cache Compression, it doesn't just chat—it acts. It searches your codebase, edits files precisely, and scaffolds entire projects without ever sending a single line of your code to the cloud. - 🛡️ Bulletproof Tool Execution: A highly fortified, single-pass regex parser catches broken XML, malformed JSON, and markdown fallbacks. If the model meant to run a tool, Nedster executes it. - 🧠 Amnesia Correction: If the LLM forgets it has filesystem access and apologizes, Nedster dynamically intercepts the refusal, injects a system correction, and forces a retry. - 🗜️ 256K Context on 8GB VRAM: Integrated with Google's TurboQuant. Feed Nedster massive log files and entire codebases without triggering out-of-memory errors. - 📚 Built-in Local RAG: Comes with a standalone ChromaDB engine to vectorize and semantically search your project directories effortlessly. - 🔁 Iteration Budgets: Hard limits on autonomous loops and continuity watchdogs ensure the model stays on track and never hangs your terminal. It takes just a few commands to get your local coding agent up and running. Windows: git clone https://github.com/unrealumanga/Nedster.git cd Nedster setup.bat start.bat Linux / Mac: git clone https://github.com/unrealumanga/Nedster.git cd Nedster chmod +x setup.sh start.sh ./setup.sh ./start.sh Once running, just type your request naturally: > fix the auth logic in src/auth.py to use the requests library > scaffold a new React project in the frontend/ folder > /stats Nedster includes a beautiful Terminal UI (TUI) that provides real-time tracking of your CPU RAM, ChromaDB vectors, and GPU VRAM polling (via nvidia-smi ), keeping you in total control of your hardware.
Genesis Park 편집팀이 AI를 활용하여 작성한 분석입니다. 원문은 출처 링크를 통해 확인할 수 있습니다.
공유