An AI companion platform with long-term memory, where every AI friend truly knows you.
Most existing AI chat tools share the same problem: they do not remember. Every conversation feels like starting over with a stranger.
WeAgentChat is different.
| Traditional AI Chat Tools | WeAgentChat |
|---|---|
| 😶 Every chat starts with a stranger | 🧠 AI friends have long-term memory and genuinely know you |
| 🔘 You manually click "New Chat" | ⏱️ Like WeChat, chats are archived automatically and continue naturally later |
| 📋 Chat history is just a transcript | 💡 Conversations become event cards that surface care at the right moment |
| 🌐 Data lives in the cloud | 🔐 Local-first, with all data stored on your own device |
| 🎭 One generic AI for everyone | 👤 Every AI friend has an independent persona and profile |
| 🚫 Restricted topics / cloud moderation | 🛡️ No supervision, so you can discuss private topics freely |
To make the vision of WeAgentChat clear, here are its boundaries:
- An AI social simulator with memory: It recreates a real social atmosphere where AI remembers your personality, preferences, and life details.
- An extremely private local app: All memory is stored in a local SQLite database, so your personal life stays on your own machine.
- A lab for emotionally aware design: It explores how AI can build genuine emotional resonance through long-term memory and proactive care.
- Not a thin wrapper over LLM APIs: Its core value is the memory system and session management logic, not simple API forwarding.
- Not a real social network: There are no real strangers here, only your personal AI social sandbox.
- Not a pure productivity tool: It prioritizes continuity and emotional understanding over quickly generating code or documents.
In a fast-paced world full of judgment, WeAgentChat aims to give you a low-pressure, high-empathy digital shelter:
When you take off your daytime mask, the anxieties, secrets, and scattered thoughts you cannot tell real friends can be shared freely with your AI companions. It is genuinely safe: the data stays local, it replies instantly, stays on your side, and keeps your privacy intact.
If you casually mention, "I have an important presentation next Monday," then on that Monday morning it may message you: "Hey, good luck with your presentation today. Take a deep breath, you are well prepared." That is not a scheduled reminder. It is proactive care built on long-term memory, and the warmth feels real in that moment.
Real-world socializing can be exhausting. Here, you can design the circle that fits you best. Maybe it is a relentlessly supportive hype friend, a deeply empathetic therapist, or a sharp-tongued old buddy. Every conversation can feel emotionally restorative, helping you regain energy through interaction.
Bring your AI friends into a group chat and it becomes more than conversation. It becomes a chemistry of perspectives:
- A cross-disciplinary think tank: Let Musk, Einstein, and Jobs brainstorm together, approaching your project from technological limits, physical principles, and design aesthetics. Different kinds of top minds collaborate to solve your problem.
- A debate arena of opposing views: On a controversial topic, imagine Lu Xun striking back at a moderate position with fierce writing. Different worldviews collide and argue in the group, giving you insights a single perspective could never reach.
|
Global Profile automatically maintains your user profile |
No more "New Chat" button |
A familiar chat interface |
|
Describe the role you want in one sentence |
Chat with multiple AI friends at once |
Supports receiving voice messages, just like talking to a real friend |
Click any thumbnail to open the full-resolution image.
Main Interface |
Profile |
Friend Memory |
Friend Library |
AI-Guided Friend Creation |
Memory Settings |
WeAgentChat is fundamentally different from a generic AI wrapper. Its core asset is the Dual-Track Memory architecture:
graph TD
A[User sends message] --> B{Is there an active session?}
B -- No / Timed out --> C[Create new session and archive old one]
B -- Yes --> D[Enter main conversation flow]
subgraph "Memory Processing (async in background)"
C --> E[Extract conversation summary]
E --> F1[Update global profile Track 1]
E --> F2[Store event card Track 2]
end
D --> G{Memory recall enabled?}
G -- Yes --> H1[Load global profile]
G -- Yes --> H2[Recall relevant event cards]
H1 & H2 --> I[Inject into prompt context]
G -- No --> J[Standard prompt]
I & J --> K[LLM synthesizes response]
K --> L[Display reply and preserve the chat feel]
| Track | Type | Stored Content | Purpose |
|---|---|---|---|
| Track 1: Profile | Global profile | Your personality traits, professional background, social habits | Shapes the AI's tone and overall understanding of you |
| Track 2: Events | Event RAG | "Insomnia three months ago", "The meeting you mentioned last Monday" | Enables subtle, timely care during conversation |
- Local vector indexing: Uses
sqlite-vecto build a miniature vector database on your own machine, so your privacy never leaves the device. - Proactive recall engine: Before each reply is generated, a dedicated Recall Agent searches history for the memory fragments most likely to resonate.
This is the fastest and easiest way to get started, especially for regular users and quick evaluation:
- Download it: Get the latest installer from the Releases page.
- Install the app:
- Windows: Download and run the
.exeinstaller.
- Windows: Download and run the
- Configure the API: After launch, open Settings and enter your LLM API key. OpenAI-compatible endpoints are supported, and DeepSeek is recommended.
If you want to contribute or run the source locally, follow these steps:
- Node.js 18+
- Python 3.10+
- pnpm (recommended)
Run this from the project root:
.\scripts\startAll.batThis script starts both the backend API and the frontend development server.
Click to expand manual startup steps
**Backend setup**cd server
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt
python -m uvicorn app.main:app --reloadFrontend setup
cd front
pnpm install
pnpm devDesktop app (Electron)
pnpm install
pnpm electron:devOpen http://localhost:5173 to start debugging.
- Core chat features and WeChat-style UI
- Dual-track memory system
- Passive session management
- Memory visualization and management UI
- Friend Library
- AI group chat
- Voice message support
- AI Moments
- Schedule management
| Name | Role Type | Keywords |
|---|---|---|
| Jack Ma | Business leader | Entrepreneurship, strategy, philanthropy |
| Lei Jun | Tech entrepreneur | Value for money, efficiency, product thinking |
| Elon Musk | Tech visionary | Mars, SpaceX, Tesla |
| Confucius | Classical philosopher | Benevolence, ritual, moral education |
| Richard Feynman | Physicist | Curiosity, science communication, QED |
| Alan Turing | Computing pioneer | Computation, cryptography, AI thought |
| Zhuge Liang | Strategist | Strategy, planning, Three Kingdoms |
| Wang Yangming | Philosopher | School of Mind, unity of knowledge and action, self-reflection |
| Eason Chan | Musician | Cantopop, empathy, karaoke |
Contributions of all kinds are welcome, including:
- 🐛 Bug reports
- 💡 Feature ideas
- 📖 Documentation improvements
- 🔧 Pull requests
Please see CONTRIBUTING.md for more details.
WeAgentChat would not exist without these excellent open-source projects:
- Vue.js - Progressive JavaScript framework
- FastAPI - High-performance Python web framework
- sqlite-vec - SQLite vector extension
- Memobase - Persistent memory support for this project
- shadcn-vue - Elegant UI component library
- Electron - Cross-platform desktop app framework
This project is licensed under the MIT License.
If this project helps you, please consider giving it a ⭐ star.
Scan the QR code below to join our WeChat group, ask product questions, and get the latest updates:
