Hardware isolation, not containers
Agents are isolated with real hardware virtualization for maximum security — a Firecracker or Hyper-V microVM per agent, not just a container.
Your own AI agents,
secure by design.
Maturana lets you build, operate, and trust your own AI agents. You drive everything from Codex; each agent runs in its own hardware-isolated hypervisor for maximum security.
A lean Rust core, a skill for every workflow, a built-in knowledge graph for memory, and a security model where secrets never enter the agent VM.
$ curl -fsSL https://www.maturana.sh/install.sh | bash
Bash · Firecracker + KVM. Append -s -- --firecracker to also provision the microVM agent host.
01Why Maturana
What makes it different.
Most agent frameworks are a wrapper around a model. Maturana is built around two ideas: the AI itself is the workflow, and an agent should be as isolated and auditable as any other production workload.
Secrets your agent can't reach
Pipelock keeps secrets like API keys and credit-card numbers out of agent reach, and an egress proxy controls which systems your agents talk to. Treat your agent as an adversary and lock it down.
Everything is a skill
Maturana is built on and for Codex, the premier OpenAI agent-engineering environment. Everything is a skill, from agent creation to tools — so you customise your agents with prompts and nothing else.
Agents build their own tools
An internal WASM engine lets Maturana agents build their own tools on the fly — safely sandboxed, with no ambient authority.
A knowledge graph, beyond markdown
Maturana ships with a shared knowledge graph that scales past markdown files. Agents build their own memory, so you don't have to.
Small Rust core, skills on top
Maturana is built in Rust with a modular core from the start. Skills are extensions to the core — you run only what you need.
02What it is
Secure, lean Codex-native agent orchestration for hardware-isolated AI agents you control.
Maturana is an open-source, Codex-native platform for building and running your own AI agents. You define an agent once, launch it into its own isolated virtual machine, and operate it entirely through Codex — the same tool you use to build it. Build and deploy your own skills in seconds.
- ▹Personal agents, not a chatbot. Each agent has its own identity, soul, runtime, channels, and permissions, declared in a single MATURANA.md contract.
- ▹Codex is the control plane. You create, edit, validate, and launch agents from Codex in the repo, using Maturana's bundled skills. No bespoke console to learn.
- ▹Each agent is isolated by hardware isolation, not Docker containers. Agents execute inside Hyper-V (Windows) or Firecracker (Linux) microVMs. The isolation boundary is hardware virtualisation, not a container.
- ▹Reach it where you already are. Pair an agent to Telegram, Slack, or Discord and talk to it like a person — with an interactive command menu and live model switching.
03Install
Get running in two commands.
The installer downloads the prebuilt binary, registers services, and installs the Codex skills. Then you build your first agent from Codex.
Prerequisites
An authenticated harness
Agents need at least one AI harness signed in: the Codex CLI (codex login) or Claude Code. The installer checks this and tells you what's missing.
KVM for Firecracker
Linux hosts use Firecracker microVMs, which need KVM available. Pass --firecracker to provision the microVM host and boot-time fleet recovery.
Hyper-V enabled
Windows hosts use Hyper-V as the isolation provider. Setup registers a privileged hostd task and boot services via one UAC-approved step.
Steps
LinuxFirecracker + KVM
$ curl -fsSL https://www.maturana.sh/install.sh | bash
Installs the CLI, services, and Codex skills. Add -s -- --firecracker to also provision the microVM agent host.
WindowsHyper-V native
$ irm https://www.maturana.sh/install.ps1 | iex
Installs the Windows runtime path with Hyper-V isolation, hostd, and boot services.
OperateCodex-first
$ cd ~/maturana && codex
Open Codex in the repo and ask it to create and launch an agent, or invoke /skills directly. That's the whole workflow.
04Help
Get help.
Stuck on setup, or want to compare notes on agent design? The community is the best place to start.
Read the documentation
The full README and docs/: install and your first agent, the MATURANA.md spec reference, architecture, and the Firecracker/Hyper-V operations guides.
Join the Discord
The fastest way to get unstuck. Ask setup questions, share agent patterns, and follow development with the community.
Read the source
Maturana is open source under BSD-3-Clause. Read the code, file issues, and see exactly how isolation, pipelock, and the graph work.
05About
About me.
16 years in strategy, tech, and transformation. MBB background. I architect and ship agentic AI systems for tier-1 financial services, telco, and PE — from boardroom strategy to agents in production.