Integuru – Integrate with platforms via the source code
Hey HN! We’re Alan and Richard from Integuru (YC W24). We generate fast, reliable integrations for platforms lacking official APIs. About 2 years ago, we released the first agent that reverse-engineers network traffic to build integrations (https://github.com/Integuru-AI/Integuru). Since then, we’ve developed a new approach to reverse-engineer platforms’ source code directly. This solution also includes authentication support. Here’s a demo: https://youtu.be/4l2L8fILC2g?si=nbWbDiFrWZIWRPM7. Many AI products need to integrate with web apps, but platforms…
What it does
In the maker’s words, at launch
Hey HN! We’re Alan and Richard from Integuru (YC W24). We generate fast, reliable integrations for platforms lacking official APIs. About 2 years ago, we released the first agent that reverse-engineers network traffic to build integrations (https://github.com/Integuru-AI/Integuru). Since then, we’ve developed a new approach to reverse-engineer platforms’ source code directly. This solution also includes authentication support. Here’s a demo: https://youtu.be/4l2L8fILC2g?si=nbWbDiFrWZIWRPM7. Many AI products need to integrate with web apps, but platforms often lack official APIs. So far, there are two main ways to integrate: browser automation and via network requests. We set out to build the original agent because we ourselves suffered from RPA’s latency, reliability, and throughput issues. The original agent solved many of the prior issues, but it wasn’t perfect either. The original agent did things the obvious way: (1) have a human do the action; (2) the agent observes the network requests and (3) recreates them. That got us far, but it only supported the path the user triggered. In production, we saw all the uncovered cases: different states, missing fields, permission differences, hidden validations, and request changes we could never catch in a single run. So we started building a new solution from the ground up. Our first step was to add agents that trigger many variations of the same action. To protect the platform’s data integrity, we added a gating layer that blocks outbound requests. This lets us observe the exact request structure, branching behavior, and platform logic without accidentally mutating the live system. But this still wasn’t enough. Some logic is hard to surface by execution alone. A lot of the business rules live in the frontend bundle. So we set out to analyze the true “answer sheet” for each platform: the source code. After experimenting, we got this working. We built a source-code analysis layer that deobfuscates and traces the code associated with each action. In practical terms, our system can handle most tricky edge cases without triggering all flows. Together, these two layers result in much better coverage of the production surface area. They support more edge cases, fail less often, and avoid a lot of the brittle one-off fixes that usually come later. Finally, we added auto-healing and API doc generation to improve reliability and the UX. We also offer a 24/7 on-call maintenance team for companies on the production plan. We now spend most of our time supporting vertical AI companies and helping them connect to their customer systems. We offer a free plan for integrating with one platform and charge for additional platforms, accounts, and overage API calls. For instance, we help healthcare AI companies connect to EHRs and payer portals, and logistics companies connect to TMSs and ERPs. Some companies are now running more than 1M monthly requests per platform. Across our production users, API calls complete in ~3 seconds at 99.9%+ success rate on average. We’re also building a library of APIs that users can use out of the box. That said, this version still has limitations we want to iterate on. Although we already tackle some anti-bot mechanisms, the agent still struggles to generate integrations with heavily anti-botted platforms. When the agent fails, our on-call team steps in to improve the agent or build the integration manually if the customer requests it. Also, the UX for generating an integration is still quite manual. Our next step is to build a CLI experience, so people and their agents can create, test, and use integrations in a much more flexible manner. This also prevents humans from having to wait for Integuru to finish its tasks. We want to one day allow developers and agents to integrate with all platforms instantly. Integuru is an ongoing effort. We’re passionate about automating integrations and would love your feedback!
Does the same job
all alternatives →
Arena Agent Mode with GitHub11d ago · arena.ai · ▲98Get real work done, moving from idea to shipping in minutes
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
Launched alongside, May 2026
the whole month →
Parallel agents, diff reviewer, and multi-model comparisons
Dev tools · May 2026 · kilo.ai
- NW
Hey HN, Henry here from Cactus. We open-sourced Needle, a 26M parameter function-calling (tool use) model. It runs at 6000 tok/s prefill and 1200 tok/s decode on consumer devices. We were always frustrated by the little effort made towards building agentic models that run on budget phones, so we conducted investigations that led to an observation: agentic experiences are built upon tool calling, and massive models are overkill for it. Tool calling is fundamentally retrieval-and-assembly (match query to tool name, extract argument values, emit JSON), not reasoning. Cross-attention…
Life & fun · May 2026 · github.com
- FM
Dev tools · May 2026 · github.com