v0.1.16

Release Notes — v0.1.16

Reasoning configuration remains user-owned

hellogrok no longer creates, migrates, reorders, or replaces reasoning_effort, reasoning_efforts, or supports_reasoning_effort for any channel. Users may configure a selected level, a reasoning menu, both, or neither; missing fields continue to use Grok Build's model catalog and provider defaults.

Recovery remains compatible with temporary reasoning projections recorded by earlier releases, so stopping or recovering the proxy can still remove those obsolete managed values without changing current user-owned settings.

Provider reasoning levels pass through unchanged

Responses, Chat Completions, and Messages now preserve provider-owned reasoning levels instead of normalizing them through a DeepSeek-specific mapping table. Unknown or future non-empty levels reach the provider unchanged and remain subject to that provider's validation.

Responses-to-Messages conversion follows Grok Build's native serializer: none and minimal do not produce Messages reasoning fields, while every other non-empty value is retained.

DeepSeek off selections keep their meaning

On the first-party DeepSeek endpoint, an explicit none selection is converted to the native thinking-off switch for Chat Completions and Messages. This also works when reasoning_effort = "none" is the only reasoning field in the model configuration; a reasoning menu is not required.

When no reasoning fields are configured, hellogrok does not add a thinking switch and leaves Grok Build and DeepSeek defaults unchanged.

---

发布说明 — v0.1.16

推理配置完全归用户管理

hellogrok 不再为任何渠道创建、迁移、重排或替换 reasoning_effort、reasoning_efforts 或 supports_reasoning_effort。用户可以配置当前档位、推理菜单、同时配置两者或全部不配置；缺失字段继续使用 Grok Build 模型目录和供应商默认行为。

恢复逻辑仍兼容早期版本记录的临时推理投影，因此停止代理或执行恢复时仍能清理这些旧受管值，同时不改动当前由用户管理的设置。

供应商推理档位保持原值

Responses、Chat Completions 和 Messages 不再通过 DeepSeek 专用映射表归一化供应商推理档位。未知或未来的非空档位会原样传给供应商，由供应商自行校验。

Responses 转 Messages 时遵循 Grok Build 原生序列化行为：none 和 minimal 不生成 Messages 推理字段，其他非空值保持不变。

DeepSeek 关闭选择保持原有语义

对于 DeepSeek 官方端点，显式选择 none 时，Chat Completions 和 Messages 都会收到原生思考关闭开关。即使模型配置中只有 reasoning_effort = "none"，无需同时配置推理菜单，也能正确关闭思考。

未配置任何推理字段时，hellogrok 不添加思考开关，保持 Grok Build 和 DeepSeek 的默认行为。