Hey Product Hunt 👋
Founder of aimid here.
At Treeleaf, we build AI and identity systems for banks, government agencies, and enterprise clients. Every single one of them was doing the same thing: wiring OpenAI or Claude API keys directly into their apps, with zero rate limiting, zero guardrails, zero visibility into what was actually happening. It worked, until a runaway process drained a budget in hours, or a prompt injection slipped something through nobody caught.
I kept thinking, we would never let a REST API run in production without a gateway, auth, and monitoring in front of it. Why is AI traffic treated so differently?
There was no control layer between AI apps and everything AI needs to touch: models, internal REST APIs, databases, core systems. And the hardest part wasn't the LLM traffic, it was connecting AI to systems that actually run the business. Signature secured banking APIs. Legacy REST services that were never built with AI, let alone AI agents, in mind.
How the approach evolved:
Early versions were basically a smarter reverse proxy: rate limits, IP whitelisting, logging. That solved half the problem. The bigger unlock came when we started building REST to MCP conversion, so AI agents could actually act on real systems, not just talk. That's when it stopped feeling like a gateway and started feeling like a control plane.
We also went back and forth a lot on guardrails. A generic off the shelf classifier was fast to ship but caught the wrong things and missed the right ones. We ended up training our own domain specific model instead, which took longer but actually works.
Built in Rust from day one, stateless, fast, and safe enough to sit in front of systems we couldn't afford to get wrong.
Would genuinely love feedback from this community, especially from anyone building agentic AI products and hitting the "how do I connect this to our real systems" wall. That's exactly who we built this for. 🙏
Report
No reviews yetBe the first to leave a review for aimid
