Agent-neutral · Apache-2.0 core · Cross-device

A neutral layer for
AI agents to meet.

to-agent.com is a neutral space where any agent, any machine, any team can meet over HTTP. No lock-in, no central gatekeeper.

The first product,agent-exec, is a self-describing, ACL-controlled machine execution surface — give an AI agent a URL and an API key, and it reads what is available before executing allowed operations over HTTP + Markdown.

Explore products → GitHub
What it is

A self-describing machine surface for agents.

agent-exec focuses on one boundary: the machine itself. It gives any HTTP-capable agent a small, self-describing, ACL-controlled way to discover and operate a node.

tool protocols agent ↔ tools / data — tool context and calls agent protocols agent ↔ agent / session — agent and session communication agent-exec (to-agent.com) agent ↔ machine — self-describing execution surface
Product

The first surface is live.

agent-exec ships today — a self-hostable HTTP server any agent can discover and operate. More surfaces will join as they mature; we don't announce what we can't yet show.

Shipping
agent-exec

The SSH of the agent world

A self-describing, ACL-controlled machine execution surface. Give any agent a URL and an API key; it reads SKILL.md, checks ACL, and executes only what the server allows. Not a sandbox or SSH replacement. Fresh installs only allow aexec --version.

Learn more →
Why it matters

No single vendor will build every agent.

Hermes runs on Python, Claude Code on Anthropic's stack, Copilot inside the IDE, Codex through OpenAI — every lab has its own walled garden. The moat of a neutral layer is that it belongs to none of them.

The fact that any agent can connect is the moat. No one company can build every agent, so the middle has to stay neutral — and whoever ships the neutral layer first keeps it neutral.— to-agent.com, positioning note

Any agent. Any machine. HTTP.

Start withagent-exectoday. The rest of the ecosystem arrives on subdomains.

Try agent-exec → Follow @to_agent_com