Research-Focused Agent Frameworks
Academic and research-oriented frameworks from leading institutions like Google Research, Stanford, and UC Berkeley, advancing the state-of-the-art in agent systems.
Autonomous research workflow that helps generate ideas, conduct literature review, write and run experiment code, create reports, and publish or retrieve work through the AgentRxiv collaborative research mechanism.
A recommender system simulator that utilizes 1,000 LLM-empowered generative agents for research into recommendation systems. Agent4Rec provides a comprehensive research environment for studying agent-based recommendation algorithms and user behavior modeling.
A multi-agent AI system built with Gemini 2.0 designed to function as a virtual scientific collaborator. AI Co-Scientist uses specialized agents (Generation, Reflection, Ranking, Evolution, Proximity, and Meta-review) to generate novel research hypotheses, conduct literature reviews, and formulate research proposals with automated feedback loops.
Sakana AI research system for automatically generating, testing, and optimizing CUDA kernels. The agentic workflow translates PyTorch operations into CUDA code and iteratively improves performance through evaluation.
Sakana AI's autonomous research agent that generates research ideas, writes experiment code, runs studies, analyzes results, and produces paper-style reports with minimal human intervention.
A research framework that introduces agents that role-play to solve tasks collaboratively through conversational dynamics. CAMEL enables agents to take different roles (user, assistant) and drive problem-solving through dialogue, with implications for training, simulations, and AI alignment research.
Self-improving coding-agent research system from Sakana AI and collaborators. The Darwin Gödel Machine rewrites its own code, evaluates variants on programming benchmarks, and archives successful improvements for open-ended exploration.
A groundbreaking research initiative that leverages advanced agent-based APIs to create self-organizing, ethically governed ecosystems of AI agents. HAAS features hierarchical control mechanisms with specialized roles including Supreme Oversight Board and Executive Agents for autonomous system governance.
Foundation action model for generalist GUI agents. OS-Atlas is trained for screen understanding and action prediction across desktop, mobile, and web interfaces, and is used for downstream computer-use agent research.
Google DeepMind research prototype for a universal multimodal AI agent that can see, hear, remember context, and respond in real time through phone and glasses-style experiences.
Altera.AL's large-scale multi-agent simulation project in Minecraft. Project Sid studies emergent social behavior from many concurrent AI agents, including cooperation, organization, and simulated community dynamics.
A research assistant that uses AI to analyze citation statements and help researchers better discover, evaluate, and understand research. Scite Assistant employs deep learning techniques to extract and classify citations based on their intent, supporting comprehensive literature reviews and combating reproducibility challenges.
An AI-powered research tool that integrates with Clarivate's Academic AI Platform to provide literature review assistance, research analytics, and metadata analysis. The platform uses curated academic data and serves over 3,000 institutions with AI agents designed for academic workflows.
Realistic benchmark and environment for evaluating autonomous web agents on reproducible tasks across self-hosted web applications. WebArena is widely used for measuring multi-step web task completion.
ServiceNow research benchmark for testing web agents on enterprise knowledge-work tasks. WorkArena evaluates agents across ServiceNow workflows such as incidents, tasks, knowledge bases, and business process navigation.
No Results Found
Try adjusting your search or filters