Reduck has raised $3M in a pre-seed round led by Daphni, a Paris venture capital firm, to help AI agents get through websites faster. These agents are software that carries out online tasks for a person, and Reduck hands them ready-made scripts, small programs that do a job on a site, instead of making them read and click through every page.
Drysdale Ventures and Galion.exe, a seed fund backed by a collective of French tech founders, also joined the round. Reduck calls itself French-American: its parent company is registered in Delaware, in the United States, and runs a subsidiary in Paris. The money goes on hiring engineers, improving the core technology and growing the script library.
Reduck’s product connects to agents such as Claude and Codex through the Model Context Protocol (MCP), an open standard for plugging tools into AI assistants. Instead of reading a page and deciding where to click, the agent calls a ready-made script that does the job directly. The company sums up the idea as “AI at build, code at run”.
Scripts That Run In The User’s Browser
An agent can pick a script from Reduck’s library or ask for a new one in plain words, with no coding needed from the user. The AI does the costly thinking once, while the script is being built. After that, each run is ordinary code, which Reduck says finishes in milliseconds and uses no AI tokens, the units AI models bill by.
The scripts run inside the user’s own Chrome browser through an extension, so the agent can reach sites where that person is already logged in. Reduck says this keeps login details from being exposed and avoids bot detection. Every script can also be called as a web address by other software, such as the workflow tool n8n.
When a website changes and a script breaks, Reduck can send the agent back to repair it. The company says its library holds more than 1,200 ready-to-run scripts covering over 250 websites. Its public catalogue gives no total but lists sites from LinkedIn and Reddit to Amazon and Gmail.
Speed Claims And Early Users
Reduck says agents using its scripts can work three times faster, and at lower cost, on jobs like pulling invoices from several suppliers or market research on LinkedIn. In its own benchmark from September, Claude with Reduck finished three tasks 2.5 times faster overall than Claude driving the browser through Anthropic’s Claude for Chrome tool.
Downloading 20 invoices from five providers took 132 seconds with Reduck against 472 seconds without it, the widest gap in the test. Reduck ran the test itself, used its pre-made scripts for those providers and described its sample as small.
Marie Brayer, partner and founder of the venture firm Mox Venture, uses Reduck to gather bills from her software-as-a-service (SaaS) subscriptions.
“Fetching my SaaS invoices used to take five hours a month. Reduck does it for me in half an hour,” she said in a testimonial on the company’s site.
Tim Liu, head of product at BentoML, says an afternoon of copying LinkedIn Sales Navigator profiles became a two-minute export. Alexandre Grux, chief product officer at the legal software company Hyperlex, says filling prospects’ accounts with real data before an enterprise demo once took hours of manual entry and now takes minutes.
A free plan covers up to 100 script runs a month. Paid plans cost $20 and $100 a month, with custom pricing for larger companies that want shared access to outside apps and their own infrastructure.
Founders From An Earlier Web Agent
Daniel Huynh, the chief executive, and Raphaël Millet, the chief operating officer, previously co-founded Mithril Security, a Paris startup that began with open-source privacy tools for AI. Mithril later turned to agents and built LaVague, an open-source framework for web agents that perform online tasks.
H, the French AI company behind the Runner H web agent, later bought Mithril Security, and Huynh has since left it. Louis Abraham, the chief technology officer, joined the pair as Reduck’s third co-founder.
Plans After The Round
Reduck is building agents that write and maintain scripts on their own, with a focus on reliability and scale, so the catalogue can grow without people coding each one. It is also preparing a developer offer for teams building agents for tools with no API (application programming interface), such as insurance and healthcare portals or in-house customer relationship management (CRM) systems.
Further out, the startup is looking at automating office knowledge work, starting with generative engine optimisation (GEO), the work of getting brands to show up in answers from AI chatbots. Reduck has not given launch dates for either product.
Reduck now has fresh pre-seed money, founders who built an earlier web agent project and a script library it wants agents to grow and repair. Its speed figures come from its own testing and from customers it quotes.
Many websites will never give agents a clean way in. Reduck wants to build that entry once, keep it working as sites change, and let every agent reuse it.
