Agent Skills for building with Smallest AI — ultra-fast text-to-speech, speech-to-text, and voice AI. These skills follow the Agent Skills specification and work with Claude Code, Cursor, and any compatible AI coding assistant.
npx skills add smallest-inc/skills| Skill | Description |
|---|---|
| text-to-speech | Generate 44.1 kHz speech with Lightning v3.1 / v3.1 Pro — 12 languages, voice cloning, ~100ms streaming |
| speech-to-text | Transcribe audio with Pulse (38 languages, 64ms TTFT streaming) and Pulse Pro (leaderboard-ranked English) |
| llm-electron | Chat completions with Electron — OpenAI-compatible drop-in, sub-300ms TTFT, 70 languages, voice-agent tool calling, automatic prefix caching |
| voice-agents | Build and operate Atoms voice agents — create agents, outbound calls, campaigns, knowledge bases, conversation logs |
| speech-to-speech | Realtime full-duplex voice conversations with Hydra — audio in, audio out, one WebSocket, built-in barge-in |
| setup-api-key | Obtain and configure a Smallest AI API key |
All skills require a Smallest AI API key:
export SMALLEST_API_KEY="your-api-key"Get one at app.smallest.ai/dashboard → Developer → API Keys, or run the setup-api-key skill.
| Product | Current model | Deprecated |
|---|---|---|
| TTS | lightning_v3.1 (default), lightning_v3.1_pro |
lightning-v2, lightning-large |
| STT | pulse, pulse-pro |
— |
| LLM | electron |
— |
| S2S | hydra |
— |
- Python —
pip install smallestai - JavaScript/TypeScript —
npm install smallestai - Vercel AI SDK — smallest-ai-vercel-provider
- Also: LiveKit, Pipecat, n8n, and more
Full docs: docs.smallest.ai — every page is AI-readable by appending .md to its URL. MCP server: https://docs.smallest.ai/_mcp/server
MIT