Today we’re launching HTTPie AI, a new way to interact with APIs ✨
Free product preview
This initial release allows you to create requests using human language. While in preview, you can use HTTPie AI for free with an HTTPie account. Gradually, AI capabilities will expand throughout the platform.
Let’s say you want to fetch the details of the httpie/desktop repo from the GitHub API. This requires translating the intent to the corresponding HTTP request and entering it into the client’s UI. You need to have some degree of understanding of the following:
- How the API or server works (e.g., hostname, endpoints, auth)
- How HTTP works (e.g., headers, methods, body)
- How to instruct the client to send the exact request
What if instead of creating requests manually, we could tell HTTPie to “fetch repo details of httpie/desktop”? Or even in another language like Spanish, «obtén los detalles del repositorio httpie/desktop»?
Well, now you can!
Craft requests using human language
You can type your prompt directly in HTTPie’s URL field. It’s similar to typing a search query in your browser’s address bar instead of the exact URL.
Alternatively, you can open the HTTPie AI dialog from the “+” menu in the library. In the desktop app, you can also use ⌘I. There’s a difference in that when you start in the URL field, accepting the generated request will update the one in the current tab instead of creating a new one.
Then hit “Generate” to get a suggestion.
If you like the request, save it. The saved request automatically gets a short name generated by the AI. If you don’t like it, you can hit “Generate” again, which often produces a different result. You can also try editing the prompt. If the request includes any placeholders or obvious errors, you can modify it straight from the dialog before saving it (or you can do that after).
There are two strategies for writing HTTPie prompts: describing the request and describing the intent.
Describing the request
You can ask the AI to generate an HTTP request by describing its low-level components like the URL, headers, and body:
Generate an array of 3 blog posts with title, author, and summary and post it to localhost 3000 api as user agent HTTPie AI
Describing the intent
For actions involving public APIs of popular services, you can describe the high-level tasks you want to accomplish:
List GitHub users
Here is another example, where the model picks up that we mean a GitHub repo:
Fetch last release details of httpie/desktop
Feedback and accuracy
Like many generative AI applications, the accuracy of the outcome may vary depending on the prompt and other factors. Review the generated request and iterate on the prompt as needed. Higher specificity often leads to better results. You can quickly correct errors directly in the dialog before saving the request.
As you use HTTPie AI, please provide quick feedback on the suggestions via the thumb-down and thumb-up buttons to help us fine-tune accuracy.
Entering the era of AI copilots
As large language models (LLMs) gain robust generative capabilities, creative tools are evolving toward AI-powered experiences.
HTTPie is now becoming the first API development platform with AI assistance, and we’d love your feedback!
How we built it
We built this initial feature preview using OpenAI’s GPT-3 models.
In the process, we refactored the openai/openai-python library to make the default installation size ~7⨉ smaller (167M → 23M). Now it’s much more manageable to deploy OpenAI-powered applications using this library in environments with code size constraints, such as AWS Lambda.
We’ll continue improving the assistant’s accuracy and adding more capabilities throughout the HTTPie platform.
- ✉️ Thoughts? Questions? Shoot us an email at email@example.com.
- 📋 You can also help us improve by completing a quick survey.
- 👉 Follow @httpie and join our Discord community to stay up to date.
- 👩💻 We’re looking for new colleagues in engineering and other roles.
PS: We’re looking for a more personable name for our assistant 💡