A runtime is the program or service that performs one Agent Run. The Agent owns the role; the runtime executes; the Run records that attempt.

​
Local runtimes in the Agent setup form

The setup form groups them as:
  • Ready on this machine: Rudder found the default command.
  • Needs setup: the command was not found.
  • Other runtimes: readiness cannot be decided with a local command check.
Always run Test runtime chain. A found command can still be signed out or misconfigured.

​
Features differ by runtime

Model choices, sessions, Skills, transcripts, cost, Browser tools, Rudder tools, and mid-run controls depend on the runtime. Codex, Claude Code, OpenCode, and Pi support Rudder-managed control tools. Cursor and Gemini use their own adapter behavior. The interface should show only what the selected runtime supports. Switching to a fallback starts another attempt; it does not make two runtimes equivalent.

​
What runtime setup does not grant

Choosing a runtime does not automatically grant access to every organization, Issue, local file, credential, website, or external service. Those permissions still come from the current Agent, task, workspace, and connection settings. Generic process and HTTP adapters are advanced plumbing and do not appear as ordinary setup choices.

​
Quick choices

  • Already use Codex locally: choose Codex, test the chain, then run one small Chat task.
  • A command is found but the test says authentication failed: sign in through that CLI outside Rudder.
  • Need Browser control: use a supported local runtime in Desktop local-trusted mode and enable Agent Browser access.