---
sidebar_position: 1
slug: /ai-agent-framework
---

# Mycelium Agent Framework (Preview)

> **Status:** Coming in 2026 – early design subject to change.

Mycelium Agents are **private, sovereign, and distributed AI systems** that run on infrastructure you control. Instead of sending your data and logic to a centralized AI cloud, agents live **inside your own environment** and connect over the **Mycelium Network**.

This page gives a high‑level preview of the framework and how it fits into the rest of Mycelium.

## What Are Mycelium Agents?

From the marketing overview:

> *“Private, Sovereign and Distributed AI You Control.”*

Mycelium Agents are designed so that:

- **Execution stays local** – Models, logic, and memory run on your own hardware.  
- **Connectivity is peer‑to‑peer** – Agents communicate over Mycelium’s encrypted mesh.  
- **Data remains private** – Your datasets, embeddings, and tools stay inside your trust boundary.  
- **Agents are portable** – They can move with you across devices, clusters, and locations.

## Design Principles

The Agent Framework is being designed around a few core principles (see **Agent Design** and **Advantages** on the main site):

- **Inside your environment** – Each agent operates entirely within your chosen infrastructure (laptop, homelab, edge node, cluster).  
- **Mesh connectivity** – Agents talk to each other across trusted nodes using Mycelium Network, without central relays.  
- **Local data access** – Agents read and write data locally (file systems, QSFS, databases) without shipping it to a third‑party.  
- **Zero‑trust posture** – Identity, policy, and attestation are enforced continuously.  
- **Portability** – Agents follow your workflows, not a particular cloud provider.

## Relationship to the Rest of Mycelium

The Agent Framework is built on top of existing Mycelium components:

- **Mycelium Network** – Encrypted IPv6 overlay and addressing fabric between nodes.  
- **Mycelium Cloud / VDC** – Kubernetes clusters and virtual data centers hosting compute.  
- **QSFS / Storage** – Quantum‑safe, replicated storage for long‑term data and model artifacts.  
- **Hosters** – Community‑run capacity providing the underlying infrastructure.

Agents are planned to run:

- On your own devices (laptops, workstations, homelabs).  
- On edge or datacenter nodes provided by hosters.  
- Inside VDCs as part of larger application stacks.

## What You Can Expect

While the full technical spec is still evolving, you can expect:

- A **runtime** for defining and running agents on your nodes.  
- A **connectivity layer** for secure P2P communication between agents.  
- A **policy model** for where agents may run and which data they may access.  
- **Composability** with existing tools – containers, Kubernetes, storage backends, and Mycelium networking.

## Timeline and Scope

- **Launch window:** Agent Framework targeted for **H1 2026**.  
- **Preview phase:** Before launch, expect design updates, early prototypes, and integration examples.  
- **Ecosystem:** The goal is to make agents first‑class citizens of the wider Mycelium / ThreeFold ecosystem.

## How to Prepare

Although the Agent Framework itself is not yet released, you can prepare today by using the underlying components:

- **Networking:** Get familiar with **[Mycelium Network](/network)** and how overlay addressing works.  
- **Compute:** Use **[Mycelium Cloud](/cloud)** and **[VDCs](/vdc)** to deploy clusters and workloads.  
- **Capacity:** If you want to run agents on your own or community hardware, explore **[Hosters](/hosters)**.  
- **Data:** Consider how sensitive data will be stored and accessed locally (e.g. QSFS).

See **[Getting Ready for Agents](/ai-agent-framework/getting-ready)** for a more practical checklist.