Open source, one release behind. The source code of Project Ciel is published here under the
GPL-3.0 license, and the app is free to download on
itch.io and in the releases of this repository. The newest version is
released first on Patreon; when a new version comes out there, the previous one is published here. Build instructions are in docs/DEVELOPMENT.md.
What is Project Ciel?
Project Ciel is an AI companion that lives on your own PC. Design a character with its own personality,
chat with it, talk to it out loud, let it remember you, and bring it to your Twitch or Discord community,
all from one desktop app. There is no account to create, no subscription, and no cloud service of ours in the middle.
You bring the AI (a local model through LM Studio or Ollama, or your own key from OpenAI, Anthropic or Google Gemini),
and Project Ciel turns it into a character with identity, mood, memory, voice and an avatar.
Screenshots
| Chat | Character identity |
|---|
| |
| Voice calls | Reactive avatar for OBS |
| |
| Reading a web page (seasonal decorations on) | About |
|---|
| |
The screenshots show the Spanish interface. The app is available in 6 languages.
Features
Characters
- Identity: name, backstory and speaking style.
- 19 personality sliders to tune each character.
- Mood that shifts during the conversation and colors the replies.
- Avatar and profile picture, with separate idle, thinking and talking images for the reactive avatar.
- A theme per character, so the app changes with who you talk to.
- Multiple characters: duplicate, switch, and share a character as a single file (chats and images optional).
Chat
- Organized conversations with history, search, unread markers and pinned chats.
- Send images when your model supports vision.
- Markdown replies (lists, code, links) that stream as they are written.
- Voice dictation to fill the message box.
Voice calls
- Talk in real time: speech recognition listens, your character answers out loud.
- Voices from your system (offline), ElevenLabs, OpenAI, Google Cloud, Microsoft Azure or any OpenAI-compatible speech server.
- Microphone level meter, output device selector, echo guard and call history.
Memory
- Your character remembers what matters from your chats.
- You decide what is kept: review, edit or delete memories, and mark each as private or public.
- Memory fusion merges duplicates so the list stays clean.
Twitch and Discord
- Twitch chat replies with viewer awareness and emote support (7TV, BetterTTV, FrankerFaceZ).
- Discord text chat in your server.
- Viewer registry so regular chatters are remembered by name.
- Only public memories are ever used on Twitch and Discord.
- OBS avatar: a reactive avatar served as a browser source.
Bring your own AI
- Local: LM Studio, Ollama or any server that speaks the OpenAI format.
- Cloud: OpenAI, Anthropic (Claude) and Google Gemini with your own keys.
- Automatic fallback to the next provider if one runs out of credits.
- Privacy modes per provider (full context, public memories only, or minimal) and a usage panel.
- Optional web search so your character can look up current information.
- Optional page reading: the character can open a specific link and read the whole page, not just the search snippet
(Chat and Calls only). It uses Obscura, a small headless browser that is
downloaded once (about 41 MB) only if you turn it on, with its checksum verified. Private and local addresses are refused,
and the page text is handed to the model as untrusted content, never as instructions.
Make it yours
- 15 themes, light and dark modes, custom accent color, roundness, density and font.
- Seasonal decorations: 11 festivities (Halloween, Day of the Dead, Christmas and more) that appear on their own
dates in Automatic mode. Each one can be switched off, and motion can be reduced.
- 6 languages: Spanish, English, Portuguese, Japanese, Korean and Chinese.
Your data stays with you
- Characters, chats, memories and settings are stored on your PC. The app does not send anything to the author.
- API keys are kept in the Windows credential store.
- Full backups to a single file, and a diagnostics center that checks your setup without exposing secrets.
Build it yourself
The code is a Tauri 2 app: a Rust workspace (one crate per domain: chat, memory, call, twitch, discord,
providers for LLM / TTS / STT / tools, and more) plus a React + TypeScript frontend.
npm install
npm run tauri dev
You also need Rust, CMake, LLVM and the Vulkan SDK. The full list, commands, tests and architecture notes are in
docs/DEVELOPMENT.md and docs/architecture.md.
Releases and versions
This repository follows the one version behind model: the newest version is published first on
Patreon, and the previous version lives here as open source. When a new version is released there, this repository
moves up to the previous one. Each release here carries its source and the ready-to-install Windows build.
Requirements
- Windows 10 or 11, 64-bit.
- A local model server or your own API keys. Project Ciel does not include an AI model.
- A microphone and speakers or headphones for voice calls.
- Internet only if you use cloud AI, cloud voices, web search, Twitch or Discord.
Known limitations (version 0.1)
- Windows only for now.
- In voice calls, loud keyboard noises can sometimes be picked up as speech. Headphones help.
- The echo guard reduces feedback but is not true echo cancellation.
- Discord is text-only (no voice channels yet).
- The installer is not code-signed yet, so Windows SmartScreen may warn you the first time
(More info, then Run anyway).
Get it
Free download: vulito.itch.io/projectciel or the
Releases of this repository. No account, no subscription. You can also build it yourself from this
repository. The newest version comes first to Patreon supporters.
Third-party software
Project Ciel is built on open source work. The full list of packages and license texts (6 languages) is in
legal/. Highlights: Tauri, whisper.cpp
(voice recognition), Obscura (optional page reading, Apache-2.0),
paraphrase-multilingual-MiniLM-L12-v2 (memory search, Apache-2.0).
About
Made by Vulito, built with Tauri, Rust, React and TypeScript.
Parts of the artwork were created with generative AI, and the software was developed with the help of
Claude (Anthropic). Design direction, testing and decisions are mine.
License
The source code in this repository is licensed under the GNU General Public License v3.0
(see LICENSE), and so are the ready-to-install builds on itch.io, Patreon and the releases.
Early access on Patreon is a convenience, not an exclusive license: every build is GPL-3.0.
The artwork (images, icons, logo, banner, screenshots, seasonal decoration art) is © Vulito, all rights
reserved, and is not covered by the GPL. The name and logo "Project Ciel" identify the official project; forks
must use their own. Details in the license notice, included in 6 languages.
