Security

Trust boundary and local-first security model

RoboTunnel is designed so the robot-side agent owns the most sensitive state. Security is not just transport encryption; it also includes where keys live, where code executes, and what never reaches the platform.

Auth model

  • Agent-platform connections use Ed25519 challenge-response.
  • The agent uses an allowlist model by default.
  • CLI and bot actions route through platform-issued user identity, not shared robot passwords.

Explicit trust boundary

The robot-side agent owns the trust boundary. Sensitive execution and provider credentials stay on the robot, and the default allowlist model is there to make that boundary explicit instead of implicit.

Local key handling

LLM provider keys stay on the robot and are encrypted at rest. The intended operating model is that inference requests go directly from the robot-side agent to the provider.

Network stance

  • Outbound-first robot networking model
  • No requirement to expose inbound ports on the robot
  • WebRTC when possible, TCP-compatible path when needed

Open-source edge

The robot-side agent is open source and auditable. This is part of the trust story, not only a licensing choice.