The Top Marketplace Startups in YC S26
The marketplace companies in YC's Summer 2026 batch are putting AI agents into the work that search bars and filters never replaced: selling, matching, and sourcing. All five are already moving transactions, whether that is cards and rentals, restaurant bookings, job introductions, shipping containers, or GPU capacity. Here are the top marketplace startups (in no particular order) from the batch.
Palisade — Personalized sales agents for online marketplaces
Palisade (palisade.run) gives every visitor to a marketplace a dedicated sales agent. The agent knows the full catalog and the platform's own interface, navigates with its own cursor to help someone find a listing, and can carry the buyer through to checkout or a booking. It also texts and emails, remembers returning customers, and passes what it learns back to sellers as recommendations, including changes to listings and to the platform itself. Integration starts with a single line of code on an existing site.
Founder and CEO Jonathan Salama studied CS at Princeton, worked on influencer selling software at Amazon, and was a founding engineer at Sage AI Labs building early iMessage shopping agents. He also co-created Lulus TV, a YouTube channel with a million subscribers. Palisade's agents have driven more than $2.7M in GMV since the batch began, on platforms including trading card sites, vintage stores, and short-term rental operators.
TryNearby — Word of mouth at scale for local restaurants
TryNearby (trynearby.com) sells restaurants a $399 monthly subscription that brings creators who live nearby into the restaurant to film, edit, and post a review. A matching agent picks the creators and sends invites as soon as a restaurant signs up, and creators book their own slots, so owners coordinate nothing. Every creator relationship runs over iMessage through an agent that handles onboarding, invites, reminders, and reschedules at a volume of more than 10,000 texts a month. Owners get their own iMessage assistant that tracks video performance, comments, and reviews.
Ten months in, the company has more than 120 paying restaurants across Southern California, over 2,500 videos posted, and roughly 35 million views concentrated within a few miles of each business. Growth is 37% month over month with retention above 90% since November, and nearly half of new restaurants come from owner referrals. CEO Yousef Abdelfattah co-founded FaZe Clan after starting as a 14-year-old YouTuber in 2010 and helped take it public at a $1B valuation. COO Obaida Albaroudi and CTO Ahmad Ibrahim previously built Clipping Exe, short-form clipping infrastructure that tracked billions of views.
"The algorithm doesn't reward a business talking about itself. It rewards people talking about the business."
Yousef Abdelfattah, CEO of TryNearbyPluto — LinkedIn for agents
Pluto (talentpluto.com) has professionals spend ten minutes talking to a voice agent about what they have done, what they are good at, and what they want next, then turns that conversation into a structured profile of competency and intent. Users review what the agent learned and choose what to share and what stays private. On the other side, companies describe who they want to meet; Pluto searches its network and the wider market, confirms interest privately, and makes an introduction when both sides want one.
More than 27,000 professionals have given Pluto their career context, and over 50 companies use it to find and contact people, including Mercor, Warp, and Rho. CEO Sahil Seth previously built an AI interviewer used by large recruiting firms and led AI recruiting work at Bain. CPO Maki Potamianos founded and scaled an edtech marketplace in college, and CTO Tony Xavier built an open source no-code editor and has research background in voice agents and retrieval systems.
Donkey — The AI-native trading company
Donkey (donkey.trade) buys directly from Chinese factories and sells to US importers at a single delivered dollar price with duty paid. Importers upload a buying list and see landed prices including tariffs and freight; Donkey's engine packs the order into containers, adding a second one when the load requires it, and the quoted price holds once signed. The company takes title, puts its own inspectors on the factory floor before the factory is paid, and offers credit-insured net-45 terms to qualified buyers.
The sourcing runs on public US customs records, which agents parse to identify the factories actually making a part and separate them from shells and forwarders. Pricing itself is deterministic rather than model-generated. A home builder in Oakland consolidated 15 SKUs into one container and paid 38% less than buying piecemeal from distributors, and the company is starting with construction buyers. CEO Benjamin Martindale grew up in his family's industrial supply business, advised Sany and Zoomlion on selling into Western markets, and speaks Mandarin; CTO Minghao Tan built agent systems for industrial supply chains and runs Donkey's customs data work. The team is split between San Francisco and China.
OpenRelay — One endpoint for inference on any chip
OpenRelay (openrelay.inc) matches inference workloads to whatever accelerator can run them cheapest. A customer sends a container or a model and OpenRelay schedules it onto the best-fit hardware across NVIDIA, TPU, Trainium, and AMD, attaches a production endpoint, and handles routing, isolation, failover, metering, and billing across more than four clouds. The company says that cuts inference costs by as much as 20%, and there is no SDK to adopt, with a CLI and REST API instead. The other side of the network is supply: data centers, neoclouds, and owners of underutilized reserved fleets can connect capacity and take demand from OpenRelay rather than build a serving platform themselves.
The network is running in production at 100 billion tokens a week, spread across 22 physical locations in Europe, APAC, North America, and the Middle East and eight different accelerator SKUs. CEO Jaden Wang built a half-megawatt data center at 20, founded Heaviside Compute, and worked on virtualization and HPC at Voltage Park after a stint at TensorDock. CTO Prashant Patel was a founding member of Amazon Bedrock at AWS, where he took Custom Model Import from zero to multi-million ARR, and later a staff engineer at Voltage Park working on inference and orchestration software for clusters of more than 10,000 GPUs.