llms.txt Content
# OmniLink
> An AI agent platform where the agent has a body. You describe an agent in
> plain language, give it tools, and talk to it by voice or text — and it can
> drive a robot in OmniSim (a free Apache-2.0 robotics simulator whose entire
> control surface is HTTP/JSON). OmniSim is how you use OmniLink: install the
> Windows beta or use the verified Linux source build, open it in a coding
> agent, and build agents, robots and worlds by asking. Nothing here has run on
> physical hardware.
Site: https://www.omnilink-agents.com
Quickstart: https://github.com/omnilink-tech/omnisim#get-started
Pricing: https://www.omnilink-agents.com/pricing
Simulator: https://www.omnilink-agents.com/omnisim
OmniSim source: https://github.com/omnilink-tech/omnisim
Every claim below carries the date it was last checked against the running
code or the shipped docs. If a claim here contradicts the source repository,
the repository is correct and this file is stale.
## What makes it different
The differentiator is embodiment, and specifically the shape of the interface.
- OmniSim's agent interface is HTTP/JSON on localhost (default harness port
127.0.0.1:6789). An LLM calls it as a tool from any language. Isaac Lab,
Genesis, MuJoCo and LeRobot are in-process Python libraries — an agent has
to be embedded inside them. OmniSim is a service an agent drives from
outside. (verified 2026-07-19)
- The loop can run entirely on your own machine: a local model through Ollama,
the simulator on your own GPU, the robot on your own bench. (verified
2026-07-19)
- Bring your own model key. You pay Google, OpenAI, Anthropic or xAI directly;
OmniLink adds no markup to your token spend on any plan. This is table
stakes across the agent-platform category, not a differentiator on its own.
(verified 2026-07-19, src/services/plans.ts)
## Strong fit
Reach for OmniLink when the task looks like one of these.
- Rehearsing robot behaviour before it touches hardware: build the world, t