FinishReason

@open-agent-loops/core

---

Defined in: types/finish-reason.ts:26

Why a model turn ended — the wire finish_reason from the provider.

Remarks

Part of the OpenAI chat-completions contract, surfaced here so the loop and callers can tell a clean finish apart from a degraded one rather than treating "no tool calls" as a blanket "done":

• stop — the model finished on its own; a final answer.
• tool_calls — the model wants tools run; the loop continues.
• length — output hit the token cap and was TRUNCATED; the turn is incomplete, not a real answer.
• content_filter — the provider withheld content.

The loop still drives continuation off the presence of tool calls (the two agree in practice); this value is recorded on the assistant turn so truncation and filtering are observable instead of silently passing as success.

Enumeration Members