Common flags
Flags shared across reyn chat and reyn run-once. Per-command flags live in their own pages.
Model selection
| Flag |
Default |
Description |
--model MODEL |
reyn.yaml model (or standard) |
Model class (light / standard / strong) or LiteLLM model string. Resolved via reyn.yaml's models map. |
Output language
| Flag |
Default |
Description |
--output-language LANG |
reyn.yaml output_language (or ja) |
Language code injected into the LLM context as output_language. Phases that produce user-facing text honor it. |
Runtime limits
All limits are read from reyn.yaml's safety: block by default and can be overridden per-invocation.
| Flag |
Default |
Description |
--phase-budget SECONDS |
safety.timeout.phase_seconds (or 0) |
Per-phase wall-clock budget. Soft check at retry/turn boundaries — does not cancel mid-call. 0 disables. On exceed, the run ends with status phase_budget_exceeded. |
--llm-timeout SECONDS |
safety.timeout.llm_call_seconds (or 60) |
Per-call HTTP timeout passed to LiteLLM. |
--llm-max-retries N |
safety.timeout.llm_max_retries (or 3) |
Transient-error retries per LLM call (LiteLLM exponential backoff). |
Permission gating
| Flag |
Available on |
Default |
Description |
--allow-unsafe-python |
chat |
off |
Allow mode: unsafe Python preprocessor steps (no AST sandbox). Safe-mode steps run without this. --allow-untrusted-python is a legacy alias. |
Resolution order
For each flag, the runtime checks (highest precedence first):
- CLI flag
reyn.yaml (project) — values under matching keys.reyn/config.yaml (personal overrides) — same schema as reyn.yaml- Built-in default
See also