Tip-offs
•
Each week, we surface promising startups & founders through data signals.
This week, we’re spotlighting open source AI agents →
AI Agents Go Open Source
The "picks and shovels" are already free and everywhere.
The Agentic AI gold rush is starting from the bottom up, not the top down. Normally, in technology cycles, proprietary infrastructure is built first, then apps are built on top, and open source alternatives eventually democratize the entire stack. For instance, Linux followed Unix, and MySQL followed Oracle.
With AI agents, the sequence is inverted. The top 10 GitHub repos from Feb have gained over 870K stars in roughly ~20 days.
Recently, NVIDIA unveiled NemoClaw, a fully open source, hardware-agnostic enterprise agent platform. It is already being pitched to Salesforce, Cisco, and Adobe. OpenClaw just broke React's 10-year GitHub record in 60 days, while ByteDance shipped DeerFlow 2.0, an open source “Super Agent” harness, and instantly topped GitHub Trending.
This time, from core agent platforms to lightweight forks, skills frameworks, and security agents, the entire stack is exploding in open first.
AI Agent Infra Startups
High-growth startups building the infra layer for AI agents, from model evaluation to orchestration. Ranked by website visit growth over the past 3 months.
Synth AI | 532% growth in 3 months
AI model training platform that logs agent behavior, identifies failure points, and fine-tunes models to reduce failure rates. Raised a $500K Pre-Seed from Y Combinator in Dec 2024.
San Francisco, US
Fastino Labs | 241% growth in 3 months
Builds task-specific language models (TLMs) for AI agent personalization, delivering 99x faster inference than traditional LLMs. Raised a $17.5M Seed round in May 2025, led by Khosla Ventures.
Palo Alto, US
Agenta | 197% growth in 3 months
Open-source LLMOps platform that centralizes prompt management, experiment tracking, and evaluation for LLM applications. Raised a $1.08M Seed round in Jun 2024 from InReach Ventures.
Berlin, Germany
Letta | 103% growth in 3 months
Developer platform for building AI agents with persistent memory and learning capabilities, spun out of UC Berkeley's AI research lab. Raised a $10M Seed round in Sep 2024 led by Felicis.
San Francisco, US
Prediction Guard | 122% growth in 3 months
Secure AI platform for enterprises to deploy and govern LLMs behind their own firewalls with output validation, hallucination checks, and toxicity filtering. Raised a $3.7M Seed round in May 2025 from Sovereign's Capital.
Lafayette, US
AI Agent Infra Founders
Newly minted founders building the infrastructure layer for AI agents, from agent evaluation and optimization to multi-agent coordination & simulation.
Abhinav Sinha
Stanford University (CS, AI Specialization). YC W25. Ex-Stanford AI Lab researcher, ex-Citadel Securities, ex-Apple. Now co-founder and CEO of Lucidic AI, building a training platform that improves AI agents through simulation and parameterization without tuning model weights.
Stanford, US
Xiangkai Zeng
Carnegie Mellon University (MS, Intelligent Information Systems). YC X25. Ex-Google DeepMind senior engineer (5 years), built a function-calling infra for Gemini. Ex-Apple, ex-Microsoft Research. Now co-founder of Klavis AI, building MCP integration platforms for AI agents to use tools reliably at scale.
San Francisco, US
Remy Tuyeras
Macquarie University (PhD, Mathematics). Ex-MIT postdoctoral researcher (5 years), focused on ML for biology & computational systems. Now co-founder & CTO of Summoner, building multi-agent architectures & cryptographic protocols for secure, self-organizing AI systems.
Massachusetts, US
Mehdi Jamei
3x founder. UC Berkeley (PhD, Electrical Engineering). Previously Director of AI at System Inc. (6 years). Founded Beacon Labs, led Bayes Impact as Executive Director. Now founder & CEO of Veris AI, building collaborative, online-learning multi-agent systems powered by reasoning AI.
New York, US
Sandra Tang
MIT (Computer Science and Engineering). YC S25. Previously product engineer at Nooks (AI-powered prospecting), software engineer at Google, TPM at Meta. Now co-founder of AgentHub, building a simulation and evaluation engine for AI agents to learn, fail, and improve before they go live.
San Francisco, US
Carried Interest
Tweets & articles of the week
OpenAI ghosted, Cursor gets the ring
