# Tonalli AI

> Tonalli is a real-time AI agent that searches the web, reads URLs, executes code, runs Q&A on PDFs/Word/Excel, and reasons in multi-step processes — streaming at over 2,000 tokens per second. Build custom agents from templates (language tutor, sales/CS, travel planner, marketing, admin assistant, code reviewer) or from scratch. Comes with a Telegram bot, live voice calls in 33 languages, scheduled reminders, and persistent memory.

Tonalli AI is a free-to-try web app — no signup required for guest use. It's built for anyone who wants a fast, agentic AI assistant that travels with them across web, mobile, and Telegram. Custom agents have their own system prompt, default model, tools, voice, and (for signed-in users) trainable knowledge base.

## Core docs

- [Open the app](https://tonalli.cloud/app): Use Tonalli without signup as a guest, or sign in for higher limits, sync, memory, and Telegram.
- [Pricing](https://tonalli.cloud/#pricing): Free for guests (20 requests/day), free for signed-in users (200 requests/day) during the free period.
- [How it works](https://tonalli.cloud/#how): Three-step intro: open the app, pick a mode, ask anything.
- [Models](https://tonalli.cloud/#models): Four fine-tuned models with Nahuatl names — Tochtli (swift), Quetzal (balanced reasoning), Ocelotl (deep instruction), Itzli (parallel-tool agent).
- [Custom agents](https://tonalli.cloud/#custom-agents): Build multichannel assistants with their own prompt, model, tools, and trainable knowledge.
- [Telegram bot](https://tonalli.cloud/#telegram): Pair your account, then chat from Telegram with text, voice notes, photos, and documents.
- [Speed](https://tonalli.cloud/#speed): Real-time inference at over 2,000 tokens per second.
- [FAQ](https://tonalli.cloud/#faq): Common questions about pricing, files, languages, privacy, and self-hosting.

## Capabilities

- Real-time agentic chat with web search, URL reader, sandboxed JavaScript executor, and safe math calculator
- Document Q&A: PDF, DOCX, XLSX, Markdown, plain text, and source code (up to 5 MB per file)
- Live voice calls with low-latency TTS and barge-in across 33 languages
- Scheduled prompts (one-shot / daily / weekly / monthly), delivered to web, Telegram, or both
- Memory that persists across chats and platforms (web ↔ Telegram)
- Personas (saved persona = prompt + model + temperature) and custom agents (persona + tools + channels + knowledge)
- Live artifacts: HTML, SVG, and Mermaid diagrams render alongside chat
- Three UI locales: Spanish (default), Portuguese (BR), English

## Privacy & data

- Guest chats stay in the browser's local storage; never sent to a database
- Signed-in chats sync via row-level security — only the owner can read their rows
- No selling, no sharing, no training on user data
- Self-hosting is supported under an MIT-style license

## Optional

- [App entry point](https://tonalli.cloud/app)
- [Sign in](https://tonalli.cloud/login)
- [Sign up](https://tonalli.cloud/signup)
- [Sitemap](https://tonalli.cloud/sitemap.xml)
- [Robots](https://tonalli.cloud/robots.txt)