Integuru

Integuru started as a clever local tool for reverse-engineering a site's private API from a HAR file, which is a powerful way to build an integration when no public API exists. You would capture your network traffic, feed it to the agent, and get back Python code. However, the company has since pivoted to a managed, production-grade integration service with a free tier and enterprise focus, backed by Y Combinator. The original open-source agent is now labeled v0 and points to the new commercial offering.

This means the tool you can install from the repository is not the one getting the new features or the one powering millions of monthly API calls. The value has shifted from a local, bring-your-own-key agent to a hosted platform. Pick the managed service if you need a production-ready, maintained integration for business. If you just want to script a website locally, you are better off using a browser automation library directly.

The GitHub repository is an archived version. The README states it is the "earliest version" and directs users to the commercial website. The installation documented on GitHub uses Python and Poetry. The website's homepage promotes a different service installed with npm and offers a free tier with limits. The repository lacks a sandbox for the generated code, which executes locally. This means any flaws in the generated Python script run with your full user permissions.

The tool generates Python code from HAR files. It constructs a dependency graph to handle complex request chains. This is a valid approach for reverse-engineering private APIs.

Integuru's documented architecture for its v0 open-source tool is specific and legible. It parses HAR files to construct a dependency graph of network requests, which is then traversed to generate Python code. This process identifies dependencies between requests to reconstruct the necessary sequence for a target action, such as obtaining an accountId before making a download request.

The project's GitHub repository explicitly states it contains the earliest version (v0) and directs users to a commercial website for the current product, which is a managed service. This makes the open-source code a demonstration of a concept, not a maintained tool, a fact that any potential user should consider.

Integuru began as an interesting open-source project for reverse-engineering private APIs, a clever technical trick but a difficult thing to monetize directly. The bring-your-own-key model was a clear signal of a pre-revenue, venture-backed experiment. Now, they've pivoted to a full-service, managed platform with a proper free tier and enterprise pricing, backed by Y Combinator. This is the search for a real business model, moving from a tool for hackers to a service for businesses.

The core risk is the fragility of relying on non-public APIs; a target platform's UI change can break the integration. This makes their 'On-call maintenance' a core, and costly, part of the value prop. The most likely exit is an acquisition by a larger integration platform like MuleSoft or a data aggregator like Plaid, who would value the team's expertise in handling these brittle, high-maintenance connections. Position: a useful service for non-critical workflows, but be aware the underlying tech is inherently fragile.

The demo shows a clever way to reverse-engineer private APIs from HAR files. However, building integrations against undocumented, internal endpoints is inherently fragile. A minor frontend change by the target platform could break our automation without warning, creating a support burden. The original open-source tool requires local Python setup and user-provided OpenAI keys, which complicates cost tracking and security review for sixty engineers.

The commercial website mentions a free tier and a CLI, but lacks pricing for team use, SSO, audit logs, or a data retention policy. This is a tool for an individual, not a team workflow. It is not ready for procurement.

The idea is solid: feed it a HAR file from your browser, and it generates Python code to replay the necessary API calls, even building a dependency graph. I like that it thinks about dependencies instead of just replaying a sequence. It runs locally, uses my OpenAI key, and the v0 source is on GitHub. The poetry install setup is straightforward.

But that's where local control ends. The GitHub repo is an archived v0. The real product is a proprietary SaaS platform with a free tier and a login wall. The v0 code is a cool proof-of-concept, but it's not the tool being sold. It's a dead end, so you can't fix it or extend it, and you're stuck on whatever models it supported back then.