Security and data handling
CodeTether separates identity, device authorization, machine trust, transport, and provider execution so one boundary cannot quietly replace another.
Execution machine
The selected Host or Node owns project locations, provider processes, native sessions, and the execution context. Provider credentials remain endpoint-local and are not copied into website code.
CodeTether services
The product architecture limits cloud-facing roles to identity, device ownership, revocation, and bounded connection support. The current website has no connected account backend and does not receive project contents through its public pages.
Relay
Relay coordinates authorized peers and forwards opaque machine TLS records. It can observe connection metadata required to operate the service, but it does not become project, conversation, provider, or machine-trust authority.
AI providers
Codex and Claude Code process prompts and project context according to their own products, authentication, configuration, and terms. Their subscriptions and data practices are separate from CodeTether’s free device allowance.
Connection protection
- Machine pairing is explicit, uses an expiring one-time code, and requires identity confirmation.
- Reconnect authenticates pinned peer identities instead of reusing a pairing code.
- Relay enrollment is not machine trust and cannot authorize provider execution by itself.
- Transport loss never causes CodeTether to replay a prompt automatically.
Offline and revoked behavior
Durable history stays readable when a machine is offline. New work remains unavailable until current machine, project, provider, and transport checks pass. Revocation affects its exact authorization boundary rather than deleting unrelated project data.
CodeTether never treats a lost network response as permission to replay provider work automatically. When execution ownership is uncertain, inspect the conversation before starting a fresh task.