Routant launched on YC to control which apps an AI agent reaches

Giving an AI agent access to a company's apps is still an all-or-nothing choice. An agent that reads your email can read every message. An IT agent fixing one server can often reach the whole network. Routant, a Y Combinator Fall 2026 company, launched with a product that sits between the agent and the apps. The founder, Aditya Mehta, says the layer enforces the company's rules on every request. The launch page had 7 votes at the time of writing.

What the product does

Routant sits between an agent and the systems it uses. An agent reaches those systems through one MCP endpoint. Private or on-prem systems connect to Routant through an outbound-only gateway. The company does not have to open any inbound ports.

The company chooses which tools and data each agent can see. Routant gives one example on its launch page. A work assistant can read work email. Messages from friends stay invisible to it. Sensitive actions wait for a person to approve them. Disallowed actions are blocked. Every request and every decision is logged.

Why do it this way

Mehta writes that teams today stitch access together with VPNs, custom integrations, credentials and approval workflows. Many rely on prompt text or configuration files to limit what the agent can do. Those are not real security boundaries. Agents end up with too much access, which is risky. Or with too little access, which puts the work back on people.

Routant moves the rules out of the prompt. The access control sits in the gateway. The gateway checks every request. A model told to ignore its instructions cannot reach a tool the gateway has not granted. For companies building agents for other companies, each customer sets its own rules. One integration then covers many customers.

Who is building it

Mehta studied computer science at Princeton. He worked as a quantitative options trader at IMC. There he built machine-learning pricing models. He writes on the launch page that automated trading taught him three things. Software needs explicit authority to act. It needs controls on decisions that matter. And it needs a record of what happened.

Routant is his attempt to give AI agents those three things when they work inside other companies' systems. The company says it is talking to CTOs and CISOs who decide what agents can access. It asks for introductions to security leaders putting agents into production. Routant's own site, at routant.ai, offers a demo.

The launch page lists Routant under the AIOps, Cybersecurity and AI tags. The audience it names is the person inside the company who signs off on access.

Source

• Routant launch, Y Combinator