Configuration配置说明

Configuration配置说明

Start with engine capabilities, then follow its installation, model sync and workspace guide. Pi remains experimental; engine capabilities differ.先了解各引擎能力，再按目标完成安装、模型同步和工作台配置。Pi 仍为实验版，各引擎功能并不相同。

运行时与模型配置

ClawPanel 支持 OpenClaw、Hermes、DSH、OpenCode 和实验版 Pi，配置项及协议转换范围随引擎版本变化。先参考引擎与驱动选择目标，不把所有引擎视为同一套 Gateway。

常见 OpenAI 兼容配置示例

下表只列出常见填写方式，不能替代对应供应商和当前运行时版本的实际兼容性验证。保存前请确认账号权限、模型名称和 Base URL，并在控制台状态或日志中检查结果。

提供商API Base URL说明
OpenAIhttps://api.openai.com按你的账号权限和模型列表填写
DeepSeekhttps://api.deepseek.com按 DeepSeek 当前模型列表填写，保存后验证连通性
Kimihttps://api.moonshot.cn按 Moonshot 当前模型列表填写，保存后验证连通性
自定义OpenAI 兼容 API Base URL需确认接口、鉴权、流式输出和模型名与当前运行时兼容

配置步骤

  1. 打开 ClawPanel，确认运行时处于可连接状态
  2. 进入「通用 → 模型渠道」，填写渠道信息；OpenClaw 的「模型配置」仍是该引擎专属入口
  3. 填入 API Key、Base URL、模型名称
  4. 保存只更新渠道库，再点击「同步到」目标引擎；编辑渠道后需要再次同步
  5. 在目标聊天页或工作台选择模型并发送完整测试消息；配置回读成功不代表真实模型调用通过

上下文窗口和输出上限不会扩大模型服务器实际加载的容量。各引擎的 API 类型、认证方式和默认模型独立；OAuth、结构化 SecretRef 不等于普通 API Key。Web 模式的测试和调用从服务端发起。

Gateway 配置

Gateway 是 OpenClaw 的服务入口之一。实际命令和参数可能随 OpenClaw 版本调整，请优先参考当前运行时内置说明或官方仓库文档。

  • 模式：local（推荐）— 本地运行
  • 绑定：loopback（仅本机）或 lan（局域网）
  • 认证：token 模式，自动生成安全令牌

环境变量

# 指定 OpenClaw 安装位置（可选）
export OPENCLAW_HOME=~/.openclaw

# 指定 Node.js 路径（可选）
export NODE_PATH=/usr/local/bin/node

Runtime and model configuration

Configuration and protocol conversion depend on your installed OpenClaw, Hermes, DSH, OpenCode or experimental Pi version. See Engines & drivers; the engines do not share a Gateway.

Common OpenAI-compatible examples

The table shows common field formats only. Real compatibility depends on your provider account, model name and installed runtime version. Save the configuration, then verify status or logs.

ProviderAPI Base URLNotes
OpenAIhttps://api.openai.comUse the models available to your account
DeepSeekhttps://api.deepseek.comUse the current DeepSeek model name available to your account
Kimihttps://api.moonshot.cnUse the current Moonshot model name available to your account
CustomOpenAI-compatible API base URLConfirm API shape, auth, streaming and model names before relying on it

Setup flow

  1. Open ClawPanel and confirm the runtime is reachable
  2. Open General → Model channels; OpenClaw Model configuration remains specific to OpenClaw
  3. Enter the API key, base URL and model name
  4. Save the channel library, then explicitly sync to the target engine; repeat synchronization after editing
  5. Select the model in its chat or workspace and send a complete test message; configuration readback is not proof of a successful model call

Configured context and output limits do not increase the capacity loaded by the model server. API types, authentication and defaults remain target-specific; OAuth and structured SecretRef are not ordinary API keys. Web-mode tests and calls originate on the server.

Gateway

Gateway is one of the OpenClaw service entry points. Commands and flags may change between runtime versions, so prefer runtime help output and official repository docs.

qt.cool