Purpose and Significance

ClawTalk is a terminal-based UI that transforms how developers interact with multiple LLM providers from the command line. Built on TypeScript and React (via Ink), it bridges local development environments with a remote OpenClaw server, enabling secure, multi-model AI conversations without exposing API keys on client machines. The project fills a critical gap for developers who need lightweight, keyboard-driven access to Claude, GPT, DeepSeek, Gemini, and other models—all from a single interface with real-time cost tracking and rate limit monitoring.

Key Features

Multi-model switching: Toggle between LLM providers (Anthropic, OpenAI, DeepSeek, Google, Moonshot) with a single keypress or slash command; model health probing ensures availability before use.
Cost and rate limit tracking: Real-time visibility into daily spend, weekly totals, monthly estimates, and per-session costs; subscription plan progress bars for services like Anthropic Max.
Persistent conversation management: Save, name, and organize chat sessions server-side with full transcript searchability and browsable history.
Voice I/O support: Push-to-talk speech input with live volume metering and text-to-speech output (requires SoX and gateway support).
Live voice chat mode: Real-time conversation capability with multiple providers including OpenAI, Cartesia, ElevenLabs, and Deepgram.
Message pinning and scheduled jobs: Pin important responses for reference and configure recurring prompts via cron scheduling.
Tailscale VPN integration: Zero-config remote connectivity with VPN status monitoring and automatic server discovery via 100.x.x.x addresses.
Model mismatch detection: Warns when gateway silently routes to an unintended model, critical for reproducible workflows.

Architecture

ClawTalk operates as a two-part system: the terminal client (this repository) runs locally on your machine, while ClawTalkGateway—an OpenClaw plugin—runs on your server. All API keys remain server-side, and the client communicates via HTTP over Tailscale VPN or local network. This design keeps credentials secure while enabling seamless remote access from any device.

Getting Started

Prerequisites

Node.js 20+
ClawTalkGateway plugin running on your OpenClaw server
Tailscale (for remote use) or local network access

Installation

Install globally via npm:

npm install -g @jokim1/clawtalk

Or build from source:

git clone https://github.com/jokim1/ClawTalk.git
cd ClawTalk
npm install
npm run build
npm link

Local Setup (Same Machine)

If running ClawTalk on the same machine as your OpenClaw server, zero configuration is required—just run clawtalk. The client defaults to http://127.0.0.1:18789, matching the gateway's default port. Optionally set a preferred model:

clawtalk config --model deepseek/deepseek-chat

Remote Setup (Different Machine)

Step 1: Configure Tailscale on both server and client machines. Install from tailscale.com/download, log in with the same account, and note your server's Tailscale IP (e.g., 100.85.123.45).

Step 2: Configure ClawTalk with your gateway URL, auth token, and optional default model:

clawtalk config --gateway http://100.85.123.45:18789
clawtalk config --token your-auth-token
clawtalk config --model deepseek/deepseek-chat

Step 3: Run clawtalk and start chatting. The status bar displays connection health (GW:● for gateway, TS:● for Tailscale).

Essential Keyboard Shortcuts

Shortcut	Action
Ctrl+T	View saved conversations
Ctrl+A	Open model picker
Ctrl+N	Start new chat
Ctrl+H	Browse message history
Ctrl+P	Voice input (push-to-talk)
Ctrl+V	Toggle text-to-speech
Ctrl+X	Exit

Supported Models and Pricing

ClawTalk supports models across multiple providers with transparent per-token pricing displayed in the UI:

DeepSeek: DeepSeek Chat ($0.14/$0.28), DeepSeek Reasoner ($0.55/$2.19)
Anthropic: Claude Opus 4.5 ($15/$75), Sonnet 4.5 ($3/$15), Haiku 3.5 ($0.80/$4)
OpenAI: GPT-5.2 and other supported models
Google: Gemini family
Moonshot: Kimi

Switch models via model picker (Ctrl+A), slash command (e.g., /model sonnet), or alias (e.g., deep, opus, gpt).

Who This Is For

Terminal-first developers: Those who prefer keyboard-driven workflows and avoid GUIs.
Multi-model researchers: Teams evaluating different LLM providers and comparing costs/performance in real time.
DevOps and SRE engineers: Need secure remote access to AI tools without exposing API keys client-side.
Cost-conscious organizations: Want granular visibility into per-session and monthly LLM spend.
Voice-enabled workflows: Projects requiring speech input/output alongside traditional chat.

Resources

GitHub Repository: jokim1/ClawTalk
Gateway Plugin: ClawTalkGateway (see repository for OpenClaw setup instructions)
Documentation: README and CLAUDE.md in the repository root
Related: Mission Control Dashboard – OpenClaw Agent, Tailscale, React Ink

Built by Claude Opus 4.5, Claude Opus 4.6, and Joseph Kim. Last updated February 2025.

Read original