How Firecrawl supports 1.5M+ users with 1 person and a Chatbase agent

86%

AI resolution rate

1.5M+

users supported

full-time support staff

At a glance

Firecrawl is building the library for superintelligence by turning the open web into clean, structured data that AI agents can read and act on. Founded in 2022, Firecrawl has grown from zero to more than 1.5 million users in three years. Backed by more than $90M of funding from Smash Capital, Nexus Venture Partners, and Y Combinator, it serves over 150,000 companies and runs one of GitHub's most-starred open-source projects. Firecrawl's customers are the teams building AI agents. So when Firecrawl needed one for its own support, it held the bar higher than almost anyone would.

“86% of the tickets that go into our Chatbase agent actually get solved, which is insane. We have 1.5 million users, and that's being serviced by one support person with Chatbase.”

Eric Ciarla, Co-founder, Firecrawl

Finding a support solution to scale with their explosive growth

In January 2026, Firecrawl's traffic spiked. More signups meant more customers, and more customers meant more tickets: billing questions, refund requests, API errors, usage limits. The standard answer is to hire a support team, but Firecrawl wanted to stay lean. That meant an AI agent needed to carry most of the volume on its own, without a large team behind it to catch what it was missing.

The evaluation: build, Pylon, Decimal, or Chatbase

Firecrawl evaluated an in-house build, Pylon, and Decimal before choosing Chatbase. Firecrawl's engineers build infrastructure for AI agents, but a support agent is a product in its own right. Building one meant pulling engineers off Firecrawl's core platform, then maintaining the integrations, evals, and edge cases. That didn't make sense for a company that was already growing super quickly. The support team then ran multiple vendors side-by-side on real tickets. Two things separated Chatbase. The first was the quality of the AI agent itself. Chatbase integrates SOTA models into its platform and provides a complete harness for giving the agent specific instructions and workflows. The second was the native integrations. Chatbase plugged straight into Firecrawl's existing systems with little engineering work.

“Our support team did a really nice job finding and testing a bunch of different tools, and Chatbase came out on top.”

Eric Ciarla, Co-founder, Firecrawl

How it works: an agent that takes action for the user

Most support bots answer questions and hand off anything that needs a system change. Firecrawl's AI agent built on Chatbase makes the change itself. Firecrawl connected Chatbase to Stripe and its internal systems, so the agent can look up accounts, read payment details, and issue refunds the way a human agent would. A billing issue gets resolved inside the conversation, end to end. Here is what that looks like on a typical ticket:

A customer writes in: they were charged this month even though they meant to cancel.
The agent pulls their account and payment history from Stripe and confirms the charge.
The agent reviews the refund policies and determines whether the criteria are met.
It issues the refund and tells the customer when to expect it.
The ticket closes. Nobody on the Firecrawl team touched it.

Built on enterprise-grade security

Giving an AI agent access to payment data required a vendor that met Firecrawl's high bar for security. Chatbase is SOC 2 Type II and GDPR compliant, with strict standards for customer data privacy. That meant Firecrawl could bring an AI agent into its support without putting at risk the trust of the more than 150,000 companies that rely on it.

“The really cool thing is they're basically able to handle sensitive payment details and things like refunds, really easily, like a human agent would do, but with AI. Which is how we can empower one person to serve a growing million-plus user base.”

Eric Ciarla, Co-founder, Firecrawl

The learning loop: resolution keeps climbing

86% resolution rate is just the beginning. When a support rep corrects an AI answer, the agent learns from it, whether it's more context about the product or how Firecrawl prefers to talk to customers. The agent also watches how humans resolve escalated tickets, then suggests updates to its own knowledge base and instructions so fewer tickets need a human next time.

“Over time the agent actually improved itself and was able to answer more customer queries based on what our human agents gave it feedback on. It really was able to learn our product and the way we like to interact with customers.”

Eric Ciarla, Co-founder, Firecrawl

Eric's advice to founders: run the test yourself

Eric's advice to other high-growth startups is the same process they followed. Connect the agent to everything, give it two weeks, and compare it against the alternatives on real tickets.

“Give it a shot. Give it access to your data, give it access to your docs, give it access to your Slack, and give it two weeks. See how many support tickets are actually solved and watch it improve over time. Do the due diligence, try different providers, but for us, Chatbase works extremely well.”

Eric Ciarla, Co-founder, Firecrawl

About

Firecrawl turns the open web into clean, structured data that AI agents can read.

Industry

AI infrastructure / developer tools

Founded

2022