Building an Autonomous Company Operating System with Crafted AI Platform MCP, Hermes and Obsidian

There is a shift happening in how we think about AI systems.

Most tools today are reactive. They wait for input, generate output, and stop there. Useful, but limited.

What if an AI system could actually operate like a living organization? One that makes decisions, executes work, learns from outcomes, and continuously improves itself over time?

That is the idea behind this architecture.

This article explores a system designed not just to assist, but to operate: a closed-loop intelligence framework powered by Crafted AI Platform Hermes, MCP agents, and Obsidian.

At the core are three components you may use,

• Crafted MCP Server — the execution layer of specialized AI agents
• Hermes — the decision engine — AI agent
• Obsidian — the memory and learning system

Together, they form a closed-loop intelligence system that can continuously plan, execute, and improve.

The Core Idea: AI as a Company Operating System

Instead of building isolated AI tools, we treat the system like a company:

• Decisions must be made
• Work must be executed
• Results must be measured
• Knowledge must accumulate
• Strategy must evolve

This creates a feedback loop:

input → decision → execution → learning → improved decision

Most systems stop at execution. This system does not.

Let's breakdown how it works.

Hermes: The Decision Layer

Hermes is the orchestration brain, it is an amazing self-improving AI agent built by Nous Research.

It’s the only agent with a built-in learning loop — it creates skills from experience, improves them during use, nudges itself to persist knowledge, searches its own past conversations, and builds a deepening model of who you are across sessions.

It does not execute tasks directly.

Instead, it:

• Interprets user input or system triggers
• Reads historical memory from Obsidian
• Determines intent (build, validate, grow, optimize)
• Creates strategy
• Breaks work into tasks
• Chooses which AI agent should execute each task

It behaves more like a assistant operating under strict constraints. Every decision must lead to execution and learning.

In this example my Hermes Agent will ask me do setup once I installed SKILL.md

Here is the location of SKILL.md

Ask hermes to install https://we-crafted.com/skills/SKILL.md. Thats it. Then you will have Crafted Company craftedcompany skill ready to use with your choice of model configured with it. Hermes will use skill and get it started.

PS. Make sure get your own CRAFTED API Key. Contact with us and have it and while installing skill Hermes Agent will ask to you.

Example after setting up Obsidian and Hermes Agent with installed skill, I have asked a question about the company data:

How do we acquire first 100 SME users?

Crafted MCP Server: The Execution

Execution steps is handled by the MCP layer (Model Context Protocol-based agents) by Crafted

The MCP server acts as a workforce of specialized agents:

• research agents
• builder agents
• growth agents
• analytics agents

Our Crafted built enterprise grade AI Agent platform exposes all of it is agents via MCP server, just call it and use with your API Key.

Each agent is stateless in terms of strategy.

They only:

• receive a task
• execute it by running our MCP server.
• return structured output
• simple and secure via platform access

The MCP server becomes the scalable execution engine of the system.

Here is the sample user logged in Crafted AI Platform Dashboard and able to see all the 100+ specialized AI agents.

Optionally you may embed anywhere, website etc you like.

To learn more visit: https://we-crafted.com

Obsidian: The Memory That Learns

The most critical part of the system is not execution — it is memory. All outcomes are stored in Obsidian as structured knowledge:

• experiments
• results
• insights
• patterns
• hypotheses
• strategy updates

Over time, system evolves and creates a compounding knowledge graph, I tested with sample company as you may see, all the information classified in CraftedCompany skill then used by skill.

Unlike traditional systems every execution becomes future intelligence This is what allows the system to improve instead of repeat.

The Closed-Loop System

The full system operates as a continuous loop:

1. Input arrives (user, cron, or event) using Hermes
2. Hermes retrieves relevant memory from Obsidian
3. Hermes defines intent and strategy
4. Work is broken into tasks
5. Tasks are sent to MCP agents
6. Agents execute and return results
7. Hermes captures outcomes
8. Insights are extracted
9. Memory is updated in Obsidian
10. Strategy is refined
11. System repeats

This loop never ends. Each cycle improves the next and you will be tightly integrated system that is evolves over time and helps operate your company, work, projects no matter context is it.

Why This Architecture Matters

Most AI systems are stateless.

They forget.

This system is different because it:

• remembers everything that matters
• learns from every execution
• updates its own strategy over time
• improves decision-making quality continuously

It behaves less like a tool and more like a growing organization.

The Key Design Shift — Crafted

The real breakthrough is separation of concerns:

• Hermes → decides
• Crafted MCP → executes
• Obsidian → remembers

This separation makes the system:

• scalable (execution can expand via agents)
• adaptive (memory evolves over time)
• controllable (Hermes enforces strategy rules)

It turns AI from a tool into an operating system for work.

Final Thought

We are moving from:

“AI that responds”

to

“AI that operates systems”

This architecture is an early step toward autonomous organizations where:

• decisions are automated
• execution is distributed
• learning is persistent
• intelligence compounds over time

The result is not just automation.

It is continuous organizational evolution powered by AI. With the Crafted AI Framework, you may integrate Hermes, Obsidian as tool they are so powerful that improve your workflow.

If you want to explore this architecture in practice contact with us, you can use the same components described in this article.

Get Access

If you want to experiment with the full Crafted ecosystem or integrate MCP-based agents into your own system:

👉 https://we-crafted.com/contact

Built by love with Crafted