integuru-ai/integuru

Integuru is a Python tool that watches the network requests your browser makes when you use a website, then writes code to reproduce those requests automatically. The goal is to let you automate actions on platforms that do not publish a public API, by figuring out what their internal API calls look like and generating runnable Python code to replicate them. The process works in two steps. First, you open a browser that Integuru controls, log into the platform you want to automate, and perform the action once by hand, such as downloading a bill or exporting a report. The browser records all the network traffic from that session into a file called a HAR file. Second, you give Integuru a short plain-English description of what you did, and an AI model (OpenAI's GPT-4o or similar) analyzes the recorded requests to figure out which one actually triggered the action and which earlier requests were needed to obtain the right IDs and tokens. The agent builds a dependency graph of those requests. Many API calls require values like an account ID or session token that come from a previous request, so Integuru traces those chains backward until it reaches the point where only the login cookies are needed. It then walks back through the graph and converts each step into a Python function, producing code you can run without touching a browser. The tool takes the HAR file and cookie file as inputs, along with the text prompt, and outputs Python code. You can also pass in variable names so the generated code accepts parameters, such as choosing which year's documents to fetch. Setup requires Python and the Poetry package manager, plus an OpenAI API key. The README recommends at least an o1-mini-level model for good results. All collected browser data stays local in files on your machine, nothing is sent anywhere beyond the OpenAI API calls during analysis.