Control and approval modes
Model source, session control, and approval policy are independent dimensions. Native model configuration does not prevent remote session control, and changing presentation detail does not change execution permission.
Use App Approvals and Agent Control when you need to review requests and session state in an interface.
Agent launch modes
| Mode | Command | Behavior |
|---|---|---|
| OriginRouter route | originrouter claude / originrouter codex | Use the configured local/cloud/remote route |
| Native configuration | originrouter claude --native-config / originrouter codex --native-config | Preserve the agent's login, model, and environment |
| Managed runtime | claude-terminal, codex-terminal | Compatibility entries for managed sessions |
Approval modes
| Mode | Behavior |
|---|---|
manual | Ask the user for supported approval decisions |
guarded | Automatically allow a conservative built-in scope |
ai_review | Ask the configured reviewer inside hard safety boundaries |
unrestricted | Skip interactive review for supported decisions |
custom | Evaluate a versioned policy document on the execution device |
Fail-closed boundaries
Unknown tools, insufficient evidence, unsafe path resolution, ambiguous shell expansion, and operations outside policy capability fall back to user review. The relay cannot replace the execution device's local authorization decision.
Select an approval mode at launch
Launch Claude with guarded approvals:
originrouter claude --originrouter-autonomy guarded
Launch Codex with AI review:
originrouter codex --originrouter-autonomy ai_review
Launch Claude with a custom policy file:
originrouter claude --originrouter-autonomy custom \
--originrouter-policy ~/.originrouter/policies/team-default.json
Presentation detail
Set activity detail to concise:
originrouter agent detail set concise
Set activity detail to standard:
originrouter agent detail set standard
Set activity detail to detailed:
originrouter agent detail set detailed