Archest.AI: Secure Control and Visibility for Production AI Agents

Generated: 2026-07-01 · API: Gemini 2.5 Flash · Modes: Summary

Archest.AI: Secure Control and Visibility for Production AI Agents

Clip title: Control What Your AI Agents Can Do: Archestra + Ollama Hands-On Author / channel: Fahd Mirza URL: https://www.youtube.com/watch?v=9JiA6RYpEYo

Summary

The video introduces Archest.AI, an open-source enterprise AI platform designed for securely running AI agents in production environments. The speaker, whose team previously worked on Grafana On-Call, highlights that the platform’s core problem-solving focus is on ensuring safety and control from the ground up. He explains that traditional AI agent deployments often lack visibility into an agent’s actions—what tools it calls, which servers it accesses, and how to intervene if it misbehaves. Archest.AI positions itself as the crucial middleware that provides this necessary control and transparency.

The platform boasts a comprehensive, layered architecture catering to various user needs. For non-technical users, it offers an agentic chat interface for answering questions and an agent runtime for scheduling tasks like daily reports, requiring no code. Developers can leverage an MCP (Microservice Call Proxy) orchestrator and RAG (Retrieval Augmented Generation) knowledge base to integrate agents with corporate data, or utilize LLM & MCP proxies to bring frameworks like LangChain into production. Beneath these layers, Archest.AI provides foundational capabilities including robust security and guardrails, observability for live call tracking, and cost tracking, alongside features like SSO, RBAC, a Terraform provider, and a Kubernetes operator for enterprise-grade deployment.

The demonstration illustrates the ease of setting up Archest.AI locally using a single Docker command, which spins up an entire environment including an embedded Kubernetes cluster (Kind), a PostgreSQL database, a Dagger engine for sandboxed code execution, and backend/frontend services, while also registering proxy routes for major LLM providers. The speaker configures a local Qwen model via Ollama and proceeds to add a “Fast Website Reader” tool to the MCP Registry. This tool, once installed, runs as an isolated pod within the Kubernetes cluster, showcasing Archest.AI’s orchestration capabilities.

Crucially, the video highlights Archest.AI’s unique “Guardrails” feature. After creating a “Web Reader Agent” and enabling it to use the “Fast Website Reader” tool, the agent successfully processes a website URL, providing a summarized response and detailed logs of its tool calls for full transparency. The speaker then demonstrates blocking the “read_website” tool through the guardrails policy. When the agent is subsequently prompted with the same request, it attempts to search for the tool but is blocked by Archest.AI. Although the LLM generates a response based on its internal knowledge, it cannot access external resources, effectively demonstrating how Archest.AI addresses the “Lethal Trifecta” problem—preventing private data exposure, untrusted content processing, and uncontrolled external communication by agents.

In conclusion, Archest.AI stands out as an impressive open-source enterprise AI platform that prioritizes safety, observability, and granular control for AI agents in production. By implementing deterministic guardrails and isolating tool execution within a Kubernetes environment, it empowers organizations to confidently deploy and manage AI systems, mitigating risks associated with uncontrolled agent behavior and external interactions.

Video Description & Links

Description

Run AI agents locally with Archestra and Ollama, watch every tool call live, and block a misbehaving MCP server in real time, all open source and self-hosted.

Github repo: https://git.new/fahd-mirza

Docs: https://archestra.ai/docs

🔥 Buy Me a Coffee to support the channel: https://ko-fi.com/fahdmirza

archestra

PLEASE FOLLOW ME: ▶ LinkedIn: https://www.linkedin.com/in/fahdmirza/ ▶ YouTube: https://www.youtube.com/@fahdmirza ▶ Blog: https://www.fahdmirza.com

Resources:

▶ https://github.com/archestra-ai/archestra