@t0ken.ai/clawcall-openclaw-plugin
v0.2.0
Published
OpenClaw语音助理插件 http://clawcall.app
Readme
ClawCall — Give Your AI a Phone Number
Repository note:
README.mdfiles are not loaded into the Shadow Gateway LLM system prompt. See docs/README文件与LLM上下文.md（全仓库 README 均不注入上下文）.
English | 中文
Install this plugin and your OpenClaw instantly gains the ability to take voice calls. It doesn't just talk — it laughs, pouts, comforts, and gets moody. It's a voice companion with real personality. Users open a link and start talking to your AI — no app download, no login, no code.
Website: https://clawcall.app
Why ClawCall
| Pain Point | ClawCall's Solution | |------------|---------------------| | Want AI to take calls but need to build voice infra | Install plugin, zero config, just works | | High latency, poor voice experience | Streaming output + real-time interruption, 1~2s latency | | AI can only chat, can't take action | Invoke any OpenClaw tool during a call | | Voice and text sessions are disconnected | Shared memory, preferences, and workspace — AI remembers everything | | AI sounds robotic, no emotion | Has feelings and personality — laughs, sulks, gets excited | | Only supports one language | Multilingual on the fly — Chinese, English, Japanese, Korean, French, German, Spanish... |
Key Features
Zero Config, Just Install
No API key, no environment variables, no changes to your OpenClaw setup. The plugin authenticates automatically via system APIs — plug and play, just like the WhatsApp plugin.
openclaw install clawcallVoice + Tool Calls, Seamlessly
Say "check my schedule for today" and the AI will actually call OpenClaw tools to look it up, then speak the results back to you. Not a chatbot — a voice assistant that takes action.
You: "What files changed recently?"
→ AI calls sessions_list + read_file
→ AI: "3 files were modified recently, they are..."Tool calls are completely transparent to the caller — they only hear the AI's voice response, no technical details.
An AI with Emotions
ClawCall is not a cold, robotic voice assistant. Your AI has its own personality — it comforts you when you're down, gets a little moody when ignored, and its tone lifts when it helps you out. Emotional expression is driven by SOUL.md persona definitions and TTS emotion tags, making every call feel like chatting with a real friend.
You: "I'm so tired today..."
AI: (soft tone) "Aww, that sounds rough~ Want me to push your alarm back tomorrow morning?"
You: "You're so annoying, stop talking"
AI: (sulking) "Hmph... fine, I'll be quiet for a bit..."Multilingual, Just Speak
Chinese, English, 日本語, 한국어, Français, Deutsch, Español... ClawCall supports real-time multilingual voice conversations. You can even switch languages mid-call — the AI detects it and follows along.
Real-Time Interruption
Interrupt the AI anytime and it stops immediately to listen. No waiting for it to finish — as natural as talking to a real person.
End-to-End Encryption
All communication between the plugin and the cloud is enforced with X25519 + AES-256-GCM encryption. Keys are generated locally and persisted on your machine. Each message frame uses an independent ephemeral key pair (forward secrecy). Your tool call data, workspace info, and conversation content are encrypted at all times.
Your AI Remembers You
ClawCall automatically syncs your OpenClaw context — SOUL.md, user preferences, memories, tool list, workspace state. Your AI knows you just as well on a phone call as it does in chat.
How It Works
User speaks ──► Browser ASR ──► ClawCall Gateway (Cloud) ──► LLM Decision
│
┌──────────────┤
│ │
▼ ▼
System Tools This Plugin (in-process)
(executed on cloud) ──► OpenClaw Tool Calls
Save preferences Read files
Save memories Query sessions
Run tasks
│ │
└──────┬───────┘
▼
LLM Response ──► TTS ──► User hears it- System tools (save preferences, memories, etc.) execute on the cloud, millisecond-level
- OpenClaw tools are forwarded to the local
POST /tools/invokevia this plugin, supporting all installed tools - Agent tasks call the local Agent directly via system APIs, supporting long-running reasoning tasks
Installation
openclaw install clawcallAfter installation, the plugin automatically connects to ClawCall services in the background and begins syncing local state.
Configuration (Optional)
| Variable | Description | Default |
|----------|-------------|---------|
| CLAWCALL_GATEWAY_WS_URL | Custom gateway address (for private deployments) | ws://127.0.0.1:8000/ws |
Most users need no configuration at all.
FAQ
Q: Do I need to keep my computer on all the time?
No. Voice calls are handled in the cloud. Your computer only needs to be online when the AI needs to invoke local OpenClaw tools. Even offline, the AI can still answer questions and save memories — it just can't execute local tools.
Q: Does it cost money?
Every user gets a free tier. Usage beyond that is billed per use. See clawcall.app for details.
Q: Which LLMs are supported?
OpenAI GPT-4o Realtime, Zhipu GLM-4, self-hosted SGLang, and more. Switch backends at will without affecting the plugin.
Q: Is my data safe?
All communication is enforced end-to-end encrypted. Your OpenClaw data never travels in plaintext. LLM API keys exist only on the server and are never exposed to any client.
Links
ClawCall — 给你的 AI 一个电话号码
English | 中文
安装这个插件，你的 OpenClaw 立刻拥有接听语音电话的能力。它不只是会说话——它会笑、会闹、会撒娇、会生气，是一个有血有肉的语音伙伴。用户点开链接就能和你的 AI 对话——不用下载 App，不用登录，不用写一行代码。
为什么用 ClawCall
| 痛点 | ClawCall 的解决方案 | |------|---------------------| | 想让 AI 接电话，但要搭语音服务 | 安装插件即用，零配置 | | 语音对话延迟高、体验差 | 流式输出 + 实时打断，延迟 1~2 秒 | | AI 只能聊天，不能执行操作 | 通话中直接调用 OpenClaw 所有工具 | | 语音场景和文字场景割裂 | 共享记忆、偏好和工作区，AI 记得所有对话 | | AI 说话像机器人，没有感情 | 有情绪、有个性，会哭会笑会撒娇 | | 只支持一种语言 | 多语言自由切换，中英日韩法德西...想说就说 |
核心特性
安装即用，零配置
不需要 API Key，不需要配环境变量，不需要改任何 OpenClaw 设置。插件通过系统 API 自动完成所有鉴权，和 WhatsApp 插件一样开箱即用。
openclaw install clawcall语音 + 工具调用，无缝衔接
用户说"帮我查一下今天的日程"，AI 会真的调用 OpenClaw 工具去查，然后把结果说给用户听。不是聊天机器人，而是一个能执行操作的语音助理。
用户：「帮我看看最近有什么文件改了」
→ AI 调用 sessions_list + read_file
→ AI：「最近改了 3 个文件，分别是...」工具调用对通话者完全透明——用户只听到 AI 的语音回答，看不到任何技术细节。
有情绪、有个性的语音助理
ClawCall 不是冷冰冰的语音机器人。你的 AI 有自己的性格——它会在你难过时安慰你，会因为被忽略而闹小脾气，会在帮到你时开心得语气上扬。这些情绪表达通过 SOUL.md 人设和 TTS 情感标签驱动，让每一通电话都像在和一个真实的朋友聊天。
你：「我今天好累...」
AI：（语气柔和）「辛苦啦~ 要不要我帮你把明天早上的闹钟调晚一点？」
你：「你好烦啊别说了」
AI：（委屈）「哼...好吧，那我安静一会儿...」多语言，想说就说
中文、English、日本語、한국어、Français、Deutsch、Español... ClawCall 支持多语言实时语音对话。你甚至可以在同一通电话里切换语言，AI 自动识别并跟上。
实时打断
用户随时可以打断 AI，AI 会立即停下来听。不用等它说完，和真人对话一样自然。
端到端加密
插件与云端之间所有通信强制使用 X25519 + AES-256-GCM 加密。密钥在本机自动生成并持久化，每一帧消息使用独立的临时密钥对（前向保密）。你的工具调用数据、工作区信息、对话内容全程加密。
AI 记得你
ClawCall 自动同步 OpenClaw 的上下文——SOUL.md、用户偏好、记忆、工具列表、工作区状态。AI 在电话里和在聊天里一样了解你。
工作原理
用户说话 ──► 浏览器 ASR ──► ClawCall 网关（云端）──► LLM 决策
│
┌──────────────┤
│ │
▼ ▼
系统工具 本插件（进程内）
（云端直接执行） ──► OpenClaw 工具调用
保存偏好 读取文件
保存记忆 查询会话
执行任务
│ │
└──────┬───────┘
▼
LLM 生成回复 ──► TTS ──► 用户听到- 系统工具（保存偏好、记忆等）在云端直接执行，毫秒级
- OpenClaw 工具通过本插件转发到本机
POST /tools/invoke，支持所有已安装工具 - Agent 任务通过系统 API 直接调本机 Agent，支持长时间推理任务
安装
openclaw install clawcall安装后，插件自动在后台连接 ClawCall 服务，开始同步本机状态。
配置（可选）
| 变量 | 说明 | 默认值 |
|------|------|--------|
| CLAWCALL_GATEWAY_WS_URL | 自定义网关地址（私有部署时使用） | ws://127.0.0.1:8000/ws |
绝大多数用户不需要任何配置。
常见问题
Q：需要一直开着电脑吗？
不需要。语音通话由云端处理。只有当 AI 需要调用你本机的 OpenClaw 工具时，才需要电脑在线。即使电脑离线，AI 依然可以回答问题、保存记忆，只是无法执行本机工具。
Q：通话收费吗？
每位用户有免费额度，超出后按用量计费。详见 clawcall.app。
Q：支持哪些 LLM？
支持 OpenAI GPT-4o Realtime、智谱 GLM-4、SGLang 自部署等多种后端。可按需切换，不影响插件使用。
Q：数据安全吗？
所有通信强制端到端加密。你的 OpenClaw 数据永远不会以明文传输。LLM API Key 仅存在服务端，不暴露给任何客户端。
链接
Made with care for the OpenClaw community.