SmallClaw 常见问题 SmallClaw FAQ

支持。SmallClaw 的模型请求会走你自己配置的模型服务，所以如果你接的是本地部署的模型服务，它就可以按本地 LLM 的方式工作。与此同时，权限、渠道、任务、运行历史和本地 Skills 的控制都仍然在本地 macOS 应用中完成。 Yes. SmallClaw sends model requests through the model service you configure, so if you connect a locally hosted model service, it can work as a local LLM setup. At the same time, permissions, channels, tasks, runtime history, and local skill control all remain inside the local macOS app.

换句话说，SmallClaw 支持本地 LLM 接入，但是否完全离线，取决于你接入的模型服务本身。 In other words, SmallClaw supports local LLM integration, but whether it is fully offline depends on the model service you connect.

做法很简单：先把本地 Gemma 模型以 OpenAI-compatible 的方式提供服务，再在 SmallClaw 里把这个服务当成一个本地模型源接入。以 Ollama 为例，通常可以先把服务地址设成 http://localhost:11434/v1，然后填入你实际拉起的 Gemma 模型名。 The setup is straightforward: expose the local Gemma model through an OpenAI-compatible service, then connect that service in SmallClaw as a local model source. With Ollama, you would typically point the endpoint to http://localhost:11434/v1 and use the actual Gemma model name you started.

在 SmallClaw 里，进入 Settings / AI Models，启用 Ollama (Local)，填写 endpoint 和 model 名，保存后就可以在 Agent Console 里测试。 In SmallClaw, open Settings / AI Models, enable Ollama (Local), fill in the endpoint and model name, save the settings, and then test it in the Agent Console.

• 先确认本地模型服务已经按 OpenAI-compatible chat completions 暴露出来。 First confirm that your local model service is exposed as OpenAI-compatible chat completions.
• 在 AI Models 里启用本地 provider，并填写正确的 endpoint 和 model 名。 Enable the local provider in AI Models and enter the correct endpoint and model name.
• 保存后到 Agent Console 里发一句简单请求，比如 Hello，确认返回正常。 After saving, send a simple request in Agent Console such as Hello and confirm that the response works.

如果连不上，优先检查三件事：endpoint 是否正确、model 名是否一致、以及本地服务是否真的是 OpenAI-compatible 接口。SmallClaw 会把本地 Gemma 当作日常对话模型、项目和任务的语义理解模型，以及本地优先、低延迟、隐私友好的工作模型来使用。 If it does not connect, check three things first: whether the endpoint is correct, whether the model name matches, and whether the local service is truly exposing an OpenAI-compatible interface. SmallClaw can then use local Gemma as a daily conversation model, a semantic understanding model for projects and tasks, and a local-first, low-latency, privacy-friendly working model.

SmallClaw 是一款原生 macOS 桌面 AI Agent 应用。它不是只会聊天的助手，而是可以接任务、组织流程、控制权限，并把自然语言指令转成实际行动的桌面工具。 SmallClaw is a native macOS desktop AI agent application. It is not just a chat assistant; it can take tasks, organize workflows, control permissions, and turn natural-language instructions into real actions on the desktop.

普通聊天型 AI 主要回答问题；SmallClaw 更进一步，能把回答变成执行链路。它能管理渠道、任务、工作流、Skills 和本地权限，让 AI 不只是说话，而是真正做事。 A normal chat AI mainly answers questions. SmallClaw goes further by turning answers into execution chains. It can manage channels, tasks, workflows, skills, and local permissions, so the AI does not just talk; it actually does work.

最直接的方式是先下载试用版。SmallClaw 是一个独立的 macOS 应用，下载后就可以直接打开，不需要先搭环境或安装一堆依赖。想继续了解，先从产品主页和技术指南开始。 The easiest way is to download the trial first. SmallClaw is a standalone macOS app, so after downloading it you can open it directly without setting up an environment or installing a long list of dependencies. If you want to keep going, start with the product homepage and technical guide.

不需要。普通使用只要会配置模型和渠道就可以。只有当你想进一步管理 Skills、自动化规则或更复杂的工作流时，才会更需要一点技术理解。 No. For normal use, you only need to configure the model and channels. You only need more technical understanding if you want to manage skills, automation rules, or more complex workflows.

SmallClaw 可以通过 WhatsApp、Telegram、Slack、iMessage 等多种渠道接收任务。你可以把它理解成一个统一中枢，不同入口都可以把请求送进来。 SmallClaw can receive tasks through channels such as WhatsApp, Telegram, Slack, and iMessage. You can think of it as a unified hub that accepts requests from multiple entry points.

它可以帮助整理信息、推进任务、生成回复、处理工作流、调用本地能力，以及把分散的操作组织成可持续执行的流程。更简单地说，它的目标不是只回答，而是帮你把事做完。 It can help organize information, move tasks forward, draft responses, handle workflows, call local capabilities, and turn scattered actions into a sustainable execution process. In short, its goal is not only to answer, but to help get the work done.

SmallClaw 的安全设计是本地优先的。敏感信息会尽量留在本机，权限由 macOS 和应用内机制共同控制，外来 Skills 也会先经过复审再进入运行链路。这样做的重点，是让用户始终知道它能做什么、不能做什么。 SmallClaw is designed with a local-first security model. Sensitive information stays local whenever possible, permissions are controlled by both macOS and app-level mechanisms, and external skills go through review before entering the runtime path. The goal is to keep it clear what the app can and cannot do.

你可以先试用，再根据自己的使用感受决定是否购买个人模式还是组织（公司）模式。具体授权方式和价格信息以购买页面为准。 You can try it first and then decide whether to buy the personal mode or the organization (company) mode based on your experience. License terms and pricing are shown on the purchase page.