See Chati.AI in Action
In 30 seconds
| |
|---|
| 🧰 She's an agent | Browses the web, reads and writes files, runs commands, makes plans, uses skills and MCP servers. With permission cards or in auto mode. |
| 🎭 She has a face | Real video stitched by emotion: she smiles, gets surprised, winks, gets angry. Several avatars, male and female, with more on the way, some of them uncensored. |
| 🎙️ Real-time voice | Speaks and listens on your PC, no internet needed. You can interrupt her by talking. |
| 👀 She sees you | Share your webcam or your screen and she comments on what she sees. |
| 📱 A phone | Her own WhatsApp account, her own number with a SIM, Google Meet video calls, Telegram and email. |
| 🌍 Multilingual | The app in 8 languages, and she answers in yours. |
| 🖥️ Local or cloud | Ollama, LM Studio, llama.cpp, your ChatGPT account, Gemini, Groq, Cerebras, Mistral, OpenRouter… |
| 🔁 Free, all day | Several free accounts: when one runs out of quota, she switches to the next one by herself. |
| 🔀 Model router | A free model for chatting and another one, paid or free, for coding. She switches by herself depending on what you ask. |
| 🔓 No added censorship | What she says depends only on the model and the persona you choose. Nobody in between. |
Install
🐧 On Linux, install.sh adds Chati.AI to your app menu and desktop and opens it. If the sound library (libjack) is missing, it installs it, asking for your password once.
All versions are also on the releases page. The app is not signed yet: if Windows shows a SmartScreen warning, click More info › Run anyway; if macOS won't open it, run xattr -dr com.apple.quarantine Chati.AI-osx-arm64 once.
5851547fc1df37ec64354b4709ea48a6c7a12d4c534895bf7e6a542e00d6533c Chati.AI-win-x64.zip
ff85675033c26f20931358c6b33e8808b8de6ffbcd6496f58415764ce842f70e Chati.AI-linux-x64.tar.gz
57be7f4d675e6428ad0694388202e326b36650d7350ecc3361448f39f71e52a6 Chati.AI-osx-arm64.tar.gz
- Download the file for your system and open it as shown above. Nothing to install: everything she
needs is inside.
- The wizard sets everything up the first time you open her:
- Language: English, Spanish, French, German, Portuguese (Brazil), Russian, Polish or Simplified Chinese.
- Her brain:
- On your PC: she checks your graphics card memory and RAM and recommends the largest model that
fits. Tick Install and configure automatically and she installs Ollama, downloads the model and
checks that it answers.
- On the internet: sign in with your ChatGPT account or paste a key from a provider.
- Look and voice: Lena, Marc or the light wave, and a voice you can listen to before choosing.
Only the voice you pick is downloaded.
- Ready: from there you can go straight to connecting WhatsApp, Telegram and the rest.
- Start talking. Type
/help in the chat to see the commands.
You can open the wizard again at any time from the menu. Prefer to set things up by hand? Use the model
button in the top bar: free keys for Gemini or Groq, your ChatGPT account, or your own
Ollama, LM Studio or llama.cpp, which she finds by herself.
Requirements: Windows 10 or 11 (64-bit), Linux (64-bit) or macOS on Apple Silicon (M1 or newer). For local models, a graphics card with at least 5 GB (9 GB for the 12B model); the smallest ones can also run on the processor if you have enough RAM. With cloud models any PC will do.
Everything she does
- Video characters: each character is a folder with its videos and cut points. Add your own and
they show up in Settings › Appearance.
- Light wave: an abstract, discreet avatar for when you don't want a face.
- Several avatars to choose from: Lena, Marc and the light wave, with more coming soon,
including some uncensored ones.
- Two moods: happy or angry. Offend her and even her idle gestures change, and it lasts.
- Built-in clip editor: mark where each gesture starts and ends and tag it. The model learns the
new gestures on its own.
- Compact mode (F11): folds the video away and leaves a small chat window at your side.
- Trigger any gesture by hand:
/beso (kiss), /triste (sad), /lengua (tongue out)…
- Your webcam or your screen, shared with her. A green border tells you she's watching.
- Screen watching: "tell me when the build finishes".
- Photos and files: drag images, PDF, Word, Excel or CSV into the chat.
- Real-time voice: she speaks with neural voices (language, voice and speed of your choice) and
listens while you talk. All on your PC, offline and with nothing to install: it comes built in and
runs on the CPU or the graphics card, your choice. You can interrupt her by talking and send without pressing Enter.
- Voice and video calls with her, right from the window.
She is not a chat locked inside a window: she can have her own accounts and talk to you like anyone else.
- Full WhatsApp: link a number with a QR code and she gets a real WhatsApp account. She chats,
sends photos and voice notes and answers voice and video calls, with her face and her voice.
- Google Meet video calls: she joins with her own Google account and talks on video in real time:
she sees you, hears you and answers out loud.
- Her own phone number (Windows only): with a USB modem and a SIM card, she reads the SMS she receives and
picks up calls to her number. For safety, she never sends SMS or makes calls on her own.
- Telegram: her own bot. She sends text, photos, videos of her gestures and voice notes. With
Away mode you carry her on your phone, and permission cards reach you there, with a PIN for
dangerous actions.
- Email: she reads your mail and sums it up. She always asks before sending.
- Contacts: you decide who can talk to her. Each contact gets their own tab and their own Chati,
with separate history and memory and no access to your tools.
- Internet with a real browser: search, read pages, click, fill in forms, download. Long pages
are summarized by a researcher sub-agent so they don't flood the model's memory.
- Files and commands: read, write and edit files; run commands (PowerShell on Windows, bash on
Linux and macOS).
- Plans: for long tasks she writes a plan and ticks it off step by step.
- Skills: import them straight from GitHub (
user/repo), from a .zip, a .md or a folder.
Claude, Codex, Cursor and skills.sh formats all work. Or ask her to make a new one: she ships with a
skill for creating skills.
- MCP servers: add any server by URL and switch its tools on and off with one click.
- Network tools: she connects to other computers over SSH and to Android phones over ADB, finds the
devices on your network and runs diagnostics (ping, DNS, open ports, TLS…). With permission cards, like
the rest of her tools.
- Subagents: she hands parts of a task to specialized subagents, coding included, to organize and
finish long jobs better.
- Model router: no need to pay for an expensive model for everything. Set a free or light model
for conversation and a stronger one, paid or free, for coding. A router decides on every message:
when what you ask touches code she hands over to the coder on her own, and comes back when you
change the subject. Its output is shown in purple so you always know who's talking. Also by hand
with
/programa and /chati.
- Reminders and recurring tasks: "remind me every 30 minutes to drink water", "tomorrow at 9
tell me…", "tell me a story non-stop".
- Manual: before touching anything she shows you a card with Run / Always allow / No. The
cards also reach your phone through Telegram.
- Auto: she does whatever it takes, but only inside the folders you share with her. Outside
them, a card.
- Dangerous commands need a PIN. The modem never sends or calls. Email is never sent without asking.
| On your PC (private, offline) | In the cloud |
|---|
| Ollama: the wizard can install it and download the recommended model for you | ChatGPT with your account: the subscription you already pay, no API |
| LM Studio | Gemini (AI Studio, free tier) |
llama.cpp: the app starts the server itself with your .gguf files, vision models included | Groq, Cerebras, Mistral, OpenRouter, all with free models |
| Any OpenAI-compatible server | OpenAI API or any other compatible one |
-
Hot switching: change models without losing the conversation. Star your favourites.
-
Several free accounts, one after another: add several keys or accounts for the same provider
and order them. When one runs out of quota she notices, switches to the next, tells you in the
chat and goes back to the first when it recovers. Stack free tiers and you have big models all day.
-
Reasoning effort of your choice (off, auto, low, medium, high) on thinking models.
-
You always see how full the model's memory is and how fast it's writing. The token counter
breaks the context down and, with your ChatGPT account, shows your plan's usage limits.
- Long-term memory: she saves what matters and recalls it in other conversations. You can view,
search and delete what she remembers.
- Context that never runs out: when the conversation fills up, she summarizes it herself and
carries on. Also by hand with
/compact.
- Editable personas: switch characters from the chat header or write your own in a
.md file.
Even the internal instructions of her sub-agents can be rewritten.
- Conversation history, several tabs, edit, delete or regenerate any message.
- Tray icon and start when you log in.
| Command | What it does |
|---|
/help | Help and command list |
/new | New chat, choosing who with |
/screen | Shows her your screen (/screen all, all screens) |
/programa | Hands over to the coder (/programa fix this… sends it straight to it) |
/chati | Brings her back |
/compact | Summarizes the conversation to free memory |
/beso, /triste… | Plays that gesture on the video |
Privacy and disclaimer
- With a local model nothing leaves your PC: not the text, not the voice, not the image.
Conversations, memories, keys and settings are stored on your machine.
- With a cloud provider, what you type goes to that provider under its own terms.
- Uncensored: Chati.AI answers any kind of question, with no added filters: what she says depends
only on the model and the character you choose. With uncensored models, she holds nothing back.
You decide how you use her.
- License: free to use, personal or professional, under the Chati.AI license. It includes
open-source components under their own licenses: see third-party notices.
Support the project
Chati.AI is made with love and far too many late nights. If she makes your day better, you can buy me a coffee.
