Fernfly: Natural Language to API Actions

How it works

• 1. Describe your application. Upload an existing OpenAPI or Swagger specification, or let the platform scan your website/APIs.
• 2. Train in minutes. Fernfly writes example conversations automatically, trains the model, and scores its accuracy. No data labelling and no ML knowledge needed.
• 3. Add it to your app. Copy one endpoint URL and an API key into your product. Works across web, mobile and voice interfaces.

For websites specifically, there is a shorter path: supply your URL, let Fernfly crawl and train against it, and receive an embed snippet.

Where teams use it

• Online stores — let customers shop by describing what they want instead of navigating filters and dropdowns.
• Customer support — allow your chatbot to fulfill repetitive and low-risk customer requests, like refunding small orders and changing delivery address.
• SaaS products — replace deep menu navigation with plain-English commands.
• Mobile applications — add a chat interface that performs real in-app actions.
• Smart home and voice — let users operate devices by speaking naturally.

Proven in production

Hoobert, an AI command bar for WooCommerce, built on Fernfly, to manage a store via English prompts. Published in the WordPress.org plugin directory, exposes 28 distinct store actions through a single endpoint. It was trained from a template in an afternoon. A merchant can type “refund order 62, damaged item” and Fernfly resolves it to the correct WooCommerce function call with the right parameters, typically in under 1.5 seconds.

Performance

• Median response latency of < 1 second.
• Tool-call accuracy of approximately 85% on evaluated workloads, with a live accuracy score published for every model version you train.
• Accuracy improves over time as real user conversations are folded back into training.

Privacy and data handling

Fernfly is private by default: your trained model runs against your application, and conversations are handled within that boundary. An optional cloud fallback to a third-party model provider handles queries your trained model cannot resolve. This fallback is enabled by default and can be switched off entirely from your settings if you require that no query leaves your environment. We also support running your custom-trained Fernfly models inside your own infrastructure for a fully air-gapped flow.

Pricing

Fernfly has a flat-fee pricing with a generous free tier. There is no per-conversation fee, no per-request fee, and no charge for running or querying a model you have trained. You are not billed through this listing.