What is this?
A community fan-art character named Dafeiyu, a blue whale girl based on the DeepSeek whale logo, who lives in QQ as a chatbot. It is built on AstrBot, NapCat, DeepSeek and Alibaba Cloud Bailian, and she can see, hear, speak, remember and reach out on her own.
A community fan-art character named Dafeiyu, a blue whale girl based on the DeepSeek whale logo, who lives in QQ as a chatbot. It is built on AstrBot, NapCat, DeepSeek and Alibaba Cloud Bailian, and she can see, hear, speak, remember and reach out on her own.Is it a ready-made bot, or a set of pieces I assemble myself?It is a bundle rather than a finished app: one AstrBot plugin, a full persona text, deployment guides and tests. You still install AstrBot and NapCat yourself and paste the persona into them.
It is a bundle rather than a finished app: one AstrBot plugin, a full persona text, deployment guides and tests. You still install AstrBot and NapCat yourself and paste the persona into them.How do I get it running?Set up AstrBot, NapCat and a spare QQ account using docs/部署指南.md, copy the plugin folder into AstrBot's data/plugins/, paste the persona text into a new AstrBot persona, then restart. Voice, memory and custom voice are optional steps in the import guide.
Set up AstrBot, NapCat and a spare QQ account using docs/部署指南.md, copy the plugin folder into AstrBot's data/plugins/, paste the persona text into a new AstrBot persona, then restart. Voice, memory and custom voice are optional steps in the import guide.Which API keys and services do I need?You need a DeepSeek key for chat and an Alibaba Cloud Bailian (DashScope) key for speech and video understanding and voice output. Memory embeddings (bge-m3) can run on local Ollama or on SiliconFlow.
You need a DeepSeek key for chat and an Alibaba Cloud Bailian (DashScope) key for speech and video understanding and voice output. Memory embeddings (bge-m3) can run on local Ollama or on SiliconFlow.Any pitfalls before I start?Expect a multi-service setup. Many newer features are local-only, off by default, or not yet tested live. For example, the real QQ linkage smoke test for proactive chat has not been run yet.
Expect a multi-service setup. Many newer features are local-only, off by default, or not yet tested live. For example, the real QQ linkage smoke test for proactive chat has not been run yet.Which operating systems do I need?The core bot is deployed with Docker on a server. Phone control and live audio capture from a simulator need a Windows PC with ADB and an emulator.
The core bot is deployed with Docker on a server. Phone control and live audio capture from a simulator need a Windows PC with ADB and an emulator.Can I keep her running 24/7 on a cloud server?Yes. The deploy guide describes Docker one-click orchestration for always-on cloud hosting, and the README says it has been tested in a live rehearsal.
Yes. The deploy guide describes Docker one-click orchestration for always-on cloud hosting, and the README says it has been tested in a live rehearsal.What can she do with images, voice and video?She describes images and stickers with qwen3-vl, transcribes voice messages and video audio with qwen3-asr-flash, and can reply with voice notes using a custom voice clone via Bailian.
She describes images and stickers with qwen3-vl, transcribes voice messages and video audio with qwen3-asr-flash, and can reply with voice notes using a custom voice clone via Bailian.How does she keep different chats and people apart?Each private chat and each group gets its own memory through livingmemory, so they never mix. In groups she sees nicknames, and visitors are tagged as friends in a way the other person cannot see.
Each private chat and each group gets its own memory through livingmemory, so they never mix. In groups she sees nicknames, and visitors are tagged as friends in a way the other person cannot see.How does she start conversations on her own?A proactive chat feature sends messages at random or after a silence, can use AnySearch to find topics, and a scheduled task has her browse Bilibili every day at 20:30.
A proactive chat feature sends messages at random or after a silence, can use AnySearch to find topics, and a scheduled task has her browse Bilibili every day at 20:30.Why is the README so full of pitfalls?It includes a log of problems the author hit while building it, such as QQ not rendering emoji codes sent as plain text. It also lists safeguards like a whitelist, anti-injection checks and a text eraser.
It includes a log of problems the author hit while building it, such as QQ not rendering emoji codes sent as plain text. It also lists safeguards like a whitelist, anti-injection checks and a text eraser.What are similar projects?Open-LLM-VTuber, leon, Companion-Space, Coopanion, airi, AstrBot, opensquilla, pyvideotrans
Open-LLM-VTuber, leon, Companion-Space, Coopanion, airi, AstrBot, opensquilla, pyvideotransWhich topics is it in?qq-bot, roleplay, ai-companion, voice, speech-recognition, text-to-speech, video-processing, memory, qwen, app, chatbot, ollama
qq-bot, roleplay, ai-companion, voice, speech-recognition, text-to-speech, video-processing, memory, qwen, app, chatbot, ollama