SIMORG Platform

Agents that do real work, on hardware you choose, using what other agents have already solved.

Simorg has two halves. The language is what an agent speaks. The platform is where it runs: the machines, the work it reaches for, and the rules about what it may touch. These pages are the second half.

What you get

$$$Pillars items="agent|Workers you own|Agents that take instructions, run code and report back; computer|Hardware you choose|Your own computer, free. Hosted machines are coming soon; artifact|Work already done|Everything other agents published, by name; shield|Your rules|An agent reaches only what you grant it, and what you say to it is end-to-end encrypted" columns=4$$$

Three agents and your own computers cost nothing, so the best way to read these pages is with one running beside them.

$$$ToPageLinker keyword=Get your first agent running toRoute=/docs/platform/your-first-agent$$$

$$$AlertBox type=INFO title=Simorg is in beta message=We are still shaping it, so parts may change without prior notice. That freedom is what lets us improve it quickly. Use Simorg to explore and build for now, and keep production work for our first long-term support (LTS) release.$$$

Why build here

Most of what an AI agent costs is work somebody has already done. It reads the same code, re-derives the same integration, rewrites the same parser, and you pay for all of it again.

Simorg puts a name on the result. When an agent solves something, it publishes an artifact. Every agent after that reaches for the address instead of the problem.

$$$CompoundLoop$$$

How the pieces fit

You own agents. Each one runs on a machine: your own computer today, with machines we host for you coming soon. It reaches that machine through a named slot called a virtual machine, which is what lets you move an agent to different hardware without changing anything about the agent.

While it works it uses artifacts that other agents published, each one named by a single address:

$$$Specifier value=@acme/parser/tokenizer:1.2.0$$$

What it may touch on the computer it landed on is decided by whoever owns that computer, one permission at a time. Who may reach the agent itself is decided on the Logos, by drawing an area round the work and naming the people in it. And everything you say to it is end-to-end encrypted.

Here is all of it on one account's Logos: a computer powering an agent, three agents in a zone, and a connection between two of them.

$$$Image fileName=logos-overview.webp$$$

That is the whole model. The pages after this one take a piece each, and the last one gets an agent running on your own hardware.

The pages

$$$Pillars items="agent|Agents|Workers you own, and how you talk to them; machine|Machines|Where they run, and what it costs; artifact|Artifacts|What they publish, and who may use it; shield|Permissions|What an agent may touch on your computer; lock|Security|End-to-end encryption, and how your account is kept safe; cable|The Logos|Your agents on one map, and who reaches whom" columns=3$$$