DietrichGebert/ponytail
Ponytail enables AI agents to generate more concise, efficient, and cost-effective code by enforcing a "lazy senior dev" methodology, prioritizing existing solutions and minimal implementation.
Awesome AI Agents › AI Assistants
Melmaphother/Science-Star
Science-Star is an open-source platform for building, extending and experimenting with AI agents aimed at scientific tasks. Its core is a ReAct engine extended with planning, action, memory and reflection modules, and it is built on top of the smolagents stack. The project ships a rich toolbox so agents can gather and process evidence: web search through SerpAPI, Tavily, DuckDuckGo and the Wayback Machine, crawling via Jina and crawl4ai, a PDF parser, browser use, inspectors for documents, audio and images, a retrieval-augmented generation retriever, and code execution. New tools can be added through a documented interface. Agent topology is configurable. A single-agent mode and a multi-agent mode that pairs a CodeAgent with a dedicated search agent are both provided, switched by configuration and launched through run_single_agent.py or run_multi_agent.py, with model loaders and tools swappable without touching core logic. Evaluation is a first-class concern: the repository includes data loaders and scorers for Humanity's Last Exam and GAIA, so benchmark runs start from one command, plus Streamlit dashboards for exploring datasets and analysing run outputs. The documentation covers installation, quick start and project structure, and the roadmap lists domain-specific chemistry and biology toolkits, additional agent architectures beyond ReAct, and more benchmarks. The authors report competitive results on a small-scale HLE subset using a mid-sized reasoning model. The project is intended for researchers and developers who want a ready harness for deep-research style scientific agents rather than a production service, and it is paired with a companion reading list of papers and benchmarks.
https://github.com/Melmaphother/Science-Star
Also in AI Assistants
HKUDS/DeepTutor
DeepTutor is an AI-powered personalized learning assistant that offers interactive knowledge Q&A, visualization, practice generation, and deep research capabilities to enhance education through a comprehensive and adaptive platform.
agentscope-ai/QwenPaw
QwenPaw is a personal AI assistant framework that provides local or cloud deployment, multi-channel connectivity, skill extensibility, and multi-agent collaboration for enhanced productivity and cr...
mukul975/Anthropic-Cybersecurity-Skills
A comprehensive, open-source library of 754 structured cybersecurity skills for AI agents, mapped across five prominent industry frameworks.
googleworkspace/cli
gws is a command-line tool that dynamically interfaces with Google Workspace APIs (Drive, Gmail, Calendar, Sheets, Docs, Chat, Admin) and includes AI agent skills for automation.
nanocoai/nanoclaw
NanoClaw is a lightweight, secure AI assistant platform that runs agents in isolated containers, providing a customizable alternative to OpenClaw with native Anthropic Claude Code integration and m...
assafelovic/gpt-researcher
GPT Researcher is an autonomous AI agent that conducts deep web and local research to generate detailed, unbiased research reports with citations.
screenpipe/screenpipe
Screenpipe transforms your computer into a personal, local, and private AI that records, searches, and automates based on all your digital activities.