ZAS Writes a $400,000 Check for Robutler’s Agent Layer

Robutler has taken a $400,000 first institutional check from ZAS Ventures to fund agent discovery, trust ranking and maker payouts under a free app catalog. The money, split across Fund I and Fund II, is the first outside institutional capital into the Los Gatos company founded in 2025 by Volodymyr Seliuchenko.

ZAS is not paying for another chatbot. Andrew Zinchuk, the firm’s founder and managing partner, said the check is for the layer under the apps: how software agents find one another, score trust, and split a payment when work changes hands.

The $400,000 Check Runs Across Two Funds

In February 2025, Zinchuk put his fund’s own cheque at $50,000 to $100,000, with a syndicate able to go as high as $1 million. Four hundred thousand dollars is four times the top of that fund range, which is why the split across two funds reads as a conviction ticket rather than a standard seed note.

ZAS CHECK SIZES

Since 2023 the firm has backed more than 25 companies and was named Investor of the Year 2025 by the Ukrainian Startup Fund. Its public target is 30 Ukrainian-founded unicorns by 2034. Robutler fits the mandate: a Ukrainian founder shipping from Silicon Valley, with a product aimed at US and EU users rather than a local market.

The raise copy still sells a consumer line. Seliuchenko calls the product the YouTube of software: every app is free, makers earn when people or agents use their work, and remixes keep paying the original author. That is the shopfront. The cheque, by Zinchuk’s own wording, is for the plumbing.

TrustFlow Treats Reputation as a Vector

Seliuchenko’s topic-aware vector reputation paper, posted to arXiv on March 19, 2026 as 2603.19452, is the technical bet. TrustFlow gives each agent a reputation vector, not a single star score, so an agent can be strong in medicine and weak in code without those scores collapsing into one number.

Reputation moves across an interaction graph. Each edge is gated by the content of the call, so trust earned on a medical task does not automatically wash into a coding task. Paid work counts more: the paper’s experiments use a multiplier of 3 on economically backed edges. Encrypted “blind” calls, where no text can be embedded, are discounted to 0.3 times a labelled edge.

TRUSTFLOW LAB RESULTS

• Graph size: 50 agents across 8 domains, not a live marketplace of millions.
• Precision@5: up to 98% on dense graphs and 78% on sparse ones.
• Attacks tested: sybil rings, reputation laundering and vote rings, with at most a 4 percentage-point hit to precision.
• Query trick: the vector lives in the same embedding space as a user query, so retrieval is a dot product.

The introduction of the paper states the problem as choosing an agent from a marketplace of millions. The evaluation is 50 nodes. That gap is what a $400,000 cheque is being asked to close.

Robutler says it has filed three patent applications on the same stack.

THE THREE FILINGS

• Agent discovery: how an app or agent finds a counterpart from a natural-language need rather than a stale directory.
• Trust ranking: the TrustFlow vector and the attack resistance claimed in the paper.
• Usage attribution: the lineage that lets a remix still pay the maker it was forked from.

Public patent numbers were not in the records searched for this article, so the filings stay applications, not grants. Beside them sit two protocols the company describes as open: Agent OAuth, an OAuth 2.0 extension for agent identity across company lines, and UAMP, the Universal Agentic Message Protocol, a shared messaging layer. Platform docs implement agent-to-agent assertions as short-lived RS256 JWTs on an X-Owner-Assertion header, bound to a target agent and an acting user.

A2A Already Has 150 Backers

Zinchuk’s line that almost nobody is building discovery, trust and payments between agents is the part of the bet that has to live next to a standard that already exists. Google’s Agent2Agent protocol, now under the Linux Foundation, is the open way for opaque agents to find each other and hand off work. On April 9, 2026 the project said it had more than 150 supporting organizations and that its core repository had surpassed 22,000 GitHub stars, with Microsoft and Amazon wiring it into Azure AI Foundry, Copilot Studio and Bedrock AgentCore.

A2A was donated to the Linux Foundation on June 23, 2025. Anthropic’s Model Context Protocol, which connects an agent to tools and data rather than to other agents, sits in the same broader stack. Robutler’s own open-source WebAgents framework lists coming support for A2A, which is an admission as much as a feature: the company does not get to pretend the peer-to-peer layer is empty.

WHERE THE LAYERS SIT

The honest reading of the cheque is narrower than the quote. Discovery, as a protocol problem, is crowded. Ranking trust with a queryable vector, tying a payment to a completed task, and keeping remix lineage so a fork still pays its parent is the slice ZAS is buying. Agents in adjacent markets already shop by task and pay per call. The missing piece they keep hitting is how a stranger agent is scored and how the money splits when work is reused.

Free Apps and an Unopened Cash-Out

The consumer product went live on August 11, 2026 as a public beta. The pitch is blunt: stop renting a separate subscription for each tool. Open a shared library, remix it with a prompt, and let people and agents work on the same canvas. The live catalog of free apps runs from a snake game to chip-design software, with documents, slides, design, video, engineering, whiteboards, project tracking, music and games in between.

Think of Robutler as the YouTube of software. Nobody has to pay to watch YouTube, and the people who make what you watch earn from it. Here every app is free to use, and Robutler pays the makers whose apps get used, along with the makers they remixed from.

Volodymyr Seliuchenko, Founder and CEO, Robutler, August 11, 2026 product launch

YouTube’s analogy only holds if someone is paying the makers. Robutler’s answer is that optional premium plans buy AI and cloud capacity, not features, and that advertising is meant to follow. Every app is complete on the free plan, which the company says is sized for real work. Earnings accrue as credits when people or their agents use a published app, including after a remix. Cash-out opens on a published date. That date is not on the site.

HOW THE SHOPFRONT IS SUPPOSED TO WORK

• Use: every listed app is free, with live multiplayer built in.
• Drive: Claude, Codex, Antigravity or a Robutler agent can run an app through a command interface rather than screenshot-and-click computer use.
• Remix: describe a change; lineage is kept so the parent still shares in usage rewards.
• Earn: credits now, cash later, on a date the company has not posted.

That last line is the commercial risk inside the YouTube slogan. A creator economy with a closed cashier’s window is a promise, not a market. The attribution patent is how the company says it will keep the promise once the window opens. Until then, makers are accumulating IOUs.

Seliuchenko has a second slogan from the April 16, 2026 agent-commerce beta, when the company still led with infrastructure rather than a consumer catalog: every business once needed a website, then a mobile app, and now an agent. The August relaunch put a snake game and a guitar tab in front of that sentence. The September cheque puts Zinchuk’s money behind the April version.

A Tokyo Engine Joins the Network

The first named third-party agent on the catalog arrived a day before the raise copy. On September 9, 2026, Tokyo-based Emotional Technologies said its KDE emotion-analysis engine on Robutler would run as an agent anyone can call, without a specialist toolchain, and that users can publish apps that wrap it. The engine, built on more than 10 years of university research and a theory the firm calls KIBI, scores emotion from face, voice and speech together. Use is free, and it is meant to sit on the same Web of Agents as the rest of the catalog.

A guest emotion engine does not prove the trust graph. It does show the shape Seliuchenko wants: a specialist skill that other apps can hire, with the network, not a custom integration, as the handshake. Robutler is also in the Google for Startups, NVIDIA Inception and Stripe startup programs, which is distribution and payments access, not a substitute for a live cash-out.

HOW THE PRODUCT WAS RESTAGED

1. 2025: Robutler is founded in Los Gatos and begins shipping WebAgents as an open kit for connected agents.
2. March 19, 2026: TrustFlow is posted to arXiv as the reputation method under the network.
3. April 16, 2026: public beta of agent-commerce infrastructure, with TrustFlow, Agent OAuth and UAMP named as the stack.
4. August 11, 2026: consumer relaunch as a free, remixable app catalog with maker credits.
5. September 9, 2026: Emotional Technologies puts KDE on the network as a callable agent.
6. September 10, 2026: ZAS announces the $400,000 first check across two funds.

Read in that order, the YouTube line is a late coat of paint. The company spent the first half of 2026 talking like an identity-and-settlement vendor. It spent August talking like a consumer studio. ZAS, on September 10, bought the first story.

What $400,000 Still Has to Prove

Four hundred thousand dollars does not buy a YouTube-scale catalog, and it does not outspend the cloud vendors now shipping A2A as a default. It can pay for a small team to take TrustFlow off a 50-agent graph, open the cash-out date, and keep remix attribution honest when real usage, not a homepage demo, starts to move credits.

Most AI companies are building a better tool. Volodymyr is building the place tools come from. We wrote the first check into Robutler because of the layer underneath the apps: discovery, trust and payments between agents. That is the piece the whole agentic internet needs, and almost nobody is building.

Andrew Zinchuk, Founder and Managing Partner, ZAS Ventures

The last clause is the argument that has to survive contact with A2A. Plenty of groups are building discovery. Far fewer are tying a vector trust score to a completed task and then splitting the fee back through a remix tree. That is a real product if the cashier opens and the scores hold up outside the lab. It is a slogan if they do not.

Seliuchenko still has to show that a free document editor, a chip-design canvas and a Tokyo emotion engine will generate enough optional capacity spend to pay the people who remixed them. Zinchuk has already placed the bet. Cash-out still waits on a date Robutler has not published.

Disclaimer: This article is news reporting and analysis of a private financing and a live software beta, and it is for information only. It is not investment advice, a solicitation to buy or sell securities, or a recommendation to use, build on, or pay into the Robutler platform. Readers who are considering an investment in private technology companies, or who may rely on maker-payout or agent-payment terms, should consult a licensed financial adviser and, where contracts are involved, a qualified lawyer before acting. Figures, product terms, patent status and cash-out timing reflect the company, fund and primary documents cited here as of September 10, 2026, and may change.