Hey! I’m Tarun and Zatana with my co-founder Alex. We’ve built Kampala, a man in the middle (mitm) proxy that allows you to agentically reverse engineer your existing workflow without bad browser automation or computer usage agents.
The average knowledge worker spends hours a day poring over legacy dashboards on on-prem solutions aggregating data across platforms. Current automation efforts use broken browser automations or computer-based user agents that are slow, complex, and inflexible.
Alex comes from a web reverse engineering background where he spent the last 7-8 years building integrations by hand for sneaker/ticket releases, sportsbook logins, and everything else. During that time he consulted with several companies and moved them away from browser-based infrastructure and into the application layer.
When we started Zatanna, we built in dental tech and had to deal with insurance reimbursement dashboards and legacy practice management solutions. Our superpower (as a particularly undifferentiated voice agent/front desk assistant company) was that we could integrate with almost any system requested. During that time we built extensive tooling (including Kampala) so we could push these integrations quickly.
The existing mitm proxy and tooling did not work for a few reasons:
They manipulated TLS and HTTP2 fingerprints over the wire that were detected by hardened anti-bots.
They had really bad MCPs that didn’t adequately expose essential features like scripts/replay.
They did not allow for the construction of workflows or actions given the pattern or sequence of requests.
As the tools we built became more powerful, we started using them internally to terminate conference participants, connect to external PMS systems, and interact with lazy apps. Alex even sent it to his property manager mom, who (with a lot of help) automated 2-3 hours of billing information entry into Yardi.
Since Kampala is a mitm, it is able to leverage existing session tokens/anti-bot cookies and automate things in seconds. Either use MCP by manually doing the workflow once, and have your favorite coding agent use Kampala to generate the script/API, or use an agent that directly generates the script/apis by telling you what to do.
We think the future of automation doesn’t lie in sending screenshots of webpages to LLMs, but instead using the underlying layer that computers actually understand. Excited to hear your thoughts/questions!