Pomerado, a startup from the YC F26 batch, has launched a platform that turns almost any website into an API that AI agents can use. The founders, Alan and Akshay, announced the product on Hacker News, where they said developers can build the integrations in minutes without help from the website's owner.
The pitch rests on a simple problem. AI agents are only as useful as the systems they can reach, and many of those systems have no programmatic access at all. Airlines, banks, e-commerce sites, insurance portals, government websites and older legacy systems typically lack an API or a Model Context Protocol (MCP) server. Agents in those cases fall back on browser automation or computer use, which the founders describe as slow, expensive and unreliable. Many companies have also banned computer use outright over security and compliance concerns.
The idea grew out of the founders' own experiments with personal assistants. After the OpenClaw assistant appeared, Alan and Akshay said they wanted to run much of their daily life through agents. They quickly ran into limits. When they asked their agents to book flights, the systems relied on computer use and failed at checkout. The problems grew worse with repetitive tasks, such as searching many airlines at once.
Both founders previously worked at McKinsey and Scale AI, where they helped drive agent deployments in healthcare, professional services and telecom. They said the biggest obstacle in each case was getting agents to work with legacy websites and apps. Pomerado started as a fix for personal assistant workflows, but the team soon saw wider uses for on-demand tool creation in older industries.
How does Pomerado turn a website into an API?
Developers, or their coding agents, start by giving the platform a short description of the task and answering a few questions about the workflow. The system then sends an agent to open a browser, explore the site and complete the task once, either by writing HTTP requests or a Playwright script. That first run records how to move through the site, which controls to use and how to get the result. If the process needs more information, such as a login or a clarification, the agent asks the user through MCP.
From that recording, a second agent builds a repeatable API. A separate model called Guardian checks the coding agent's actions against a set of guardrails. The company also runs the integration against synthetic inputs to confirm it has not been overfitted to one example. A final agent publishes an input and output schema.
Are login credentials safe with Pomerado's AI agents?
Sites that require sign-in and two-factor authentication are handled by a deterministic sign-in process, the company said, so the AI model and its recording tools never see customer credentials. Users can store time-based one-time password (TOTP) seeds or link their account to a Pomerado phone number, which lets the platform fill in two-factor codes automatically.
When a tool stops working, a maintenance job tries to repair it. It can fix a broken step, such as a renamed button, add features like searching business-class flights along with economy, or add new guardrails, such as requiring reservation dates to be in the future. The founders said some users have improved tools by simply asking the system to make them faster.
How fast is Pomerado compared with computer use?
The trade-off, according to the company, is that the first run takes longer. After that, the MCP tools run deterministically, which the founders say makes them much faster and cheaper than computer use. Pomerado says results usually arrive in 5 to 30 seconds. The company describes the approach as best suited to repeated, high-volume tasks.
Early users include a personal trainer who uses agents to prepare workouts and message clients. The trainer said agents could not log in and use the relevant site reliably before, so the work was done by hand. Car dealerships are connecting their Grok Bots to legacy dealer systems, including marketplaces, customer relationship management tools and enterprise resource planning software. The founders said those dealers had moved away from computer use, which could take an hour or more. Personal assistant developers are also using the platform to connect to a range of consumer websites.
Pomerado offers an open-source version under the MIT license on GitHub, along with a free hosted version. The founders said they welcome questions and feedback in the comments on their Hacker News post.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.