Personal Assistant 10 turns

In production this is a Telegram bot on AWS Lambda, built on my MAS library, wired to my real Google Calendar, Gmail, Spotify, vision-based nutrition logger, and audio transcriber. This web demo runs the same kind of LLM reasoning, with a fixed plausible state for every tool. Ask it anything.

Calendar · mocked

Gmail · mocked

Spotify · mocked

Nutrition · mocked

LLM · live

Provider chain · Groq → Gemini → NVIDIA

Try asking

In production: Telegram → AWS Lambda → MAS multi-agent orchestrator → real OAuth2 to Google Calendar, Gmail, Spotify; ElevenLabs TTS; vision model for food photos; Whisper for audio. Per-chat history persisted in S3. Calls back via the Telegram Bot API.

In this demo: Browser → /api/personal-assistant (Vercel Python serverless) → provider chain → response. The "tool state" is a fixed plausible week injected into the system prompt; the LLM reasons over it as if it had just queried the real services. No data persists.