Open Agent Summit 2025
2025-10-21
San Francisco, United States
who it's for
agent developers, machine learning engineers, research scientists, open-source maintainers, ml infrastructure engineers, security researchers, startup founders
topics
agent-to-agent protocols (A2A/AP2); multi-agent coordination and specialization; automatic agent red teaming; open-source agent platforms (AG2/AGNTCY); tool calling and MCP standardization; RL for agentic systems and SkyRL; high-performance inference with vLLM; notebook-style agents with LlamaIndex/NotebookLlama; observability and OpenTelemetry for agents; agent evaluation benchmarks and Cyber Gym; decentralized payments and identity for agents (X42/AP2)
| title | author | length |
|---|---|---|
| Keynote: Welcome & Opening Remarks | Matt White | 07m |
Conference logistics: rooms on second and third levels, PyTorch associate training across the hall, startup showcase at 5:00 in rooms 2014 and 2016 and welcome… | ||
| Keynote: Towards Building Safe and Secure Agentic AI | Dawn Song | 27m |
Automatic end-to-end red teaming frameworks generate exploitable inputs against deployed agents: agent visual uses iterative blackbox optimization to craft… | ||
| Frontiers of Agentic AI | Chi Wang | 25m |
An open-source platform AG2 (fka AutoGen) grew to tens of thousands of users and top GitHub rank within months and powers projects from Google agent platforms… | ||
| The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic... | Yutaro Yamada | 32m |
An end-to-end system runs agentic tree search with an experiment progress manager that creates parallel nodes for prototype builds, hyperparameter sweeps, main… | ||
| Building and Sharing Reliable AI Agents with BeeAI | Sandi Besen | 28m |
BeeAI is an open source framework plus platform that enforces deterministic agent rules while keeping LLM reasoning, with a Requirement Agent that limits which… | ||
| Agent2Agent Protocol and the Future of Agent Collaboration | Chitra Venkatesh | 16m |
A2A is an open standard that enables agents to discover and call each other using a JSON agent card served at a well‑known URL, communicate over HTTPS/gRPC… | ||
| Using LlamaIndex to Build an Open-Source NotebookLM Alternative | Tuana Celik | 26m |
Built an open-source NotebookLM alternative called NotebookLlama that uses LlamaIndex Workflows 1.0 and Llama Cloud (Llama par + Llama extract) to parse PDFs… | ||
| Panel Discussion: The Future of Agents and Human-Agent Collaboration | Yujian Tang, Davide Crapis, Mitesh Patel, Bradley Axen, Dawn Song | 40m |
Panel reports that agents already autonomously solve real bug bounties and can earn tens of thousands of dollars, and that the Cyber Gym benchmark finds… | ||
| Keynote: Welcome Back | Matt White | 01m |
Keynote announces a concrete PyTorch Foundation roadmap with timelines for agent-focused APIs and an open governance charter for model contributions, outlines… | ||
| Keynote: Building an Open Agentic Future | Samuel Colvin | 15m |
Pydantic AI is an open agent framework that claims fast growth and a practical alternative to closed lab platforms, with concrete choices: switch models in one… | ||
| The Model Context Protocol: Connecting AI to Everything | Adam Jones | 33m |
MCP standardizes how agents call tools by exposing tools, resources, and prompts as servers so agents swap connectors (Asana↔Jira, Google Calendar… | ||
| Accelerating Open-Source RL and Agentic Inference with vLLM | Michael Goin | 39m |
vLLM is an open-source inference engine that uses paged KV caching and continuous batching to cut KV memory overhead and boost utilization for multi-turn… | ||
| AGNTCY: The Open Source Project for Building Multi-Agent Systems | John Parello | 26m |
AGNTCY is an open source stack that stitches agents from any vendor into global multi-agent systems: identity and signed credentials, a directory like DNS for… | ||
| SkyRL: A Modular RL Library for LLM Agents | Sumanth Hegde, Tyler Griggs | 27m |
SkyRL is an open-source RL stack that trains multi-turn, tool-using LLM agents with a modular trainer/generator/controller architecture, standardizing on the… | ||
| End-to-End Optimizing Multi-Turn RL and High-Performance Inference in Agents with... | Chenyang Zhao | 32m |
SGLang presents concrete methods and numbers: an oversampling + partial rollouts (APRIL) pipeline that cuts multi‑turn RL training time by about 40%… | ||
| Panel Discussion: Agents in the Wild: Real-World Deployments and Lessons Learned | Yujian Tang, Raji Rajagopalan, Diamond Bishop, Mahdi Ghodsi, Andrew Hoh | 44m |
Hosts describe concrete deployments: a hosted MCP agent platform, AMD running models via Lemonate and CPU/GPU choices, and Microsoft building observability… | ||
| Keynote: Closing Remarks | Matt White | 01m |
Claims building an Agentic AI ecosystem rather than just a standalone technology, anchored by the PyTorch Foundation and emphasizing openness over silos and… | ||
| nothing matches | ||