Participated · Series A · Aug 2026

Sapiom provides infrastructure to operate AI agents in production, focusing on routing execution across models, compute, tools and services to optimize for cost, quality, latency, reliability and policy. Its product suite includes Sapiom Router for model selection, Sapiom Agent Studio for local development and deployment with codebase context, and Sapiom Runtime for managed execution with access controls, routing, recovery and step-level visibility. The company emphasizes reducing the cost of agent execution and enforcing budgets, permissions, and audit trails at runtime. Since launching six months ago the platform has processed over 270 million transactions and runs more than 100,000 agent executions per day, and customers have reported substantial inference cost savings. Founded in San Francisco in 2025, Sapiom has raised $50 million to date and operates its own inference capacity while expanding its team across engineering, ML, GTM and DevRel.

Participated · Series A · Apr 2026

Yuzu Health provides a vertically integrated TPA platform that handles claims processing, payments, and member administration for health plans and enables plans and brokers to launch and scale benefit designs without operational complexity. The company is led by CEO Jan‑Felix Schneider and operates across all 50 states. Yuzu raised $35M in a Series A—bringing total funding to $40M—and plans to use the capital to expand its engineering team and scale its platform nationally. The board will gain General Catalyst’s Alex Tran following the round. The company positions itself as the infrastructure behind health insurance plans and serves as a third‑party administrator to payers and brokers.

Participated · Series A · Mar 2026

RunSybil builds an AI agent called Sybil that continuously probes live applications to find, exploit and document real security vulnerabilities without humans in the loop. Its approach focuses on testing running systems—exploring, chaining vulnerabilities and testing authentication boundaries—rather than analyzing source code pre-deployment. The company was co-founded in 2023 by Ari Herbert-Voss, who previously led security research at OpenAI, and Vlad Ionescu, who led offensive security red teams at Meta. RunSybil reports working with startups (including Cursor, Turbopuffer, Notion, Baseten and Thinking Machines Lab) as well as several major financial institutions and Fortune 500 customers. Customers have reported that Sybil discovered critical vulnerabilities missed by traditional penetration tests and red teams. Investors and founders position the product as particularly relevant for regulated industries facing strict audit and compliance requirements as AI-driven development accelerates.

Participated · Equity · Feb 2026

Founded in 2023 by Kei Hayashi, Suhas Sumukh and Hardeep Gambhir, LocalHost runs 50-day, cohort-based founder labs that act as launchpads for technical and creative entrepreneurs. The company currently hosts programs in Bengaluru, Tokyo and Cluj-Napoca and is expanding to San Francisco, while also planning hardware-focused labs in Japan and France. Each Bengaluru cohort selects 15 founders who build projects ranging from AI agents and robotics to India-native language models and multidisciplinary media products. Labs provide shared workspaces, specialized hardware, operational mentorship and a community micro-fund to remove execution bottlenecks. Since opening its India lab, more than 30 venture partners and ecosystem participants have engaged with the cohorts, and alumni startups include Maya Research, Prava Payments, Dawn Labs, Whisperwave, Markov and Flashmates. The company has just secured fresh capital to broaden infrastructure, enhance hardware capabilities and scale operational teams worldwide. This $2.5 million raise positions LocalHost to onboard additional founder cohorts and deepen its global footprint.

Participated · Seed · Sep 2025

Composite, founded in early 2024 by former Uber product manager Yang Fan Yun and server-proxy entrepreneur Charlie Deane, delivers an AI agent that operates inside whatever browser a user already prefers. By installing a lightweight Mac extension, professionals can direct the agent to click, type and navigate across multiple web tools, eliminating manual work such as triaging Jira backlogs, sourcing candidates and compiling marketing reports. Because it runs in the authenticated browser session, Composite avoids separate connectors and lets administrators restrict sites or features for security. The product already surfaces task suggestions based on user behavior, and the team is developing automatic task surfacing and a scheduler for recurring workflows. Composite positions itself against single-browser AI offerings from OpenAI, Perplexity and others by focusing on power-user, cross-site professional use cases. The startup has attracted $5.6 million in seed funding to accelerate product development and prove long-term agent efficiency.