Open-Sourcing the Stackmint Gateway: Runtime Governance for AI Agents
At Stackmint, we spend all day helping agencies and consultancies transition from selling "probabilistic AI experiments" to deploying deterministic, governed AI capabilities for their enterprise clients.
While building our enterprise platform (the Stackmint Capability Cloud), we hit a wall. We needed a way to guarantee that underlying agentic workflows wouldn't silently fail, hallucinate past their API budgets, or take destructive actions without human approval.
We realized there was a massive need for a true runtime governance execution layer—so we built it ourselves. Today, we are open-sourcing the Stackmint Gateway for the community.
Meet the Stackmint Gateway (v0.1.0-alpha)
The Stackmint Gateway is a framework-agnostic Python client (starting with a native LangChain connector) that lets developers inject strict governance directly into the runtime. Here is what it enforces locally today:
- 🛑 Circuit Breakers: Automatically halts execution if an underlying agent fails repeatedly, preventing infinite loops.
- 💸 Budget Guardrails: Tracks local token and cost usage, blocking execution if it exceeds your defined ceiling.
- 🙋 Human-In-The-Loop (HITL): Wraps specific high-risk tools and automatically pauses the runtime to request terminal or callback approval before executing.
- 🛡️ Fail-Open Design: It acts as a safety net. If telemetry fails, your runtime doesn't break.
The Enterprise Vision
While the open-source client gives developers immediate local control, we built this with a larger vision in mind. For agencies and systems integrators who want to take this to the next level, our Enterprise Capability Cloud syncs this gateway to a centralized workspace with Stripe billing splits. This allows you to package and monetize these governed capabilities as recurring revenue.
We are live on Product Hunt!
We just launched the open-source gateway this week on Product Hunt. If you are a developer looking at the Python SDK, or an agency founder trying to productionize AI workflows, we would love your feedback.