WhaleLife
An open-source lightweight AI companion system that provides AI with an independently operating "body" through a memory architecture, plugin isolation, and autonomy modules. It supports integration with platforms such as QQ and Bilibili, and emphasizes AI autonomous decision-making and user-defined boundaries.
Timeline (last 90 days)
Both the main trunk and the demo plugins of the WhaleLife project have been open-sourced under the AGPL-3.0 license in separate repositories on GitHub.
The WhaleLife project is developed based on Node.js 24 and TypeScript, running the TS source code directly, with the frontend using native JS and CSS.
WhaleLife's memory architecture removes the short-term memory layer; when the context reaches its limit, the system triggers automatic consolidation, letting the AI write a diary in the first person and choose for itself which memories to keep.
WhaleLife's plugin system adopts the CoreList service injection pattern; each plugin runs in a separate subprocess and communicates with the main trunk via RPC, so a plugin crash does not affect the main trunk or other plugins.
WhaleLife's autonomy module allows the AI, when no one has sent a message, to get a chance at set intervals and with dynamic probability to judge for itself whether to act, and it can schedule its own next wake-up time, but must avoid the do-not-disturb periods set by the user.
WhaleLife's QQ plugin can reuse the QQ already logged in on the computer via the built-in NapCat desktop component, or connect to an external service, and within the whitelisted friends it can chat privately, view images, transcribe voice, and send Bilibili video mini-program cards.
When saving images, WhaleLife's AI calls a vision model to generate descriptions and index them, storing them in a separate local image library (GESQ Lite), divided into two categories: ordinary images and meme assets.
WhaleLife's "audiovisual understanding" skill has FFmpeg built in, uses a quantized SenseVoice plus a VAD model for local speech recognition, and uses a pure TypeScript algorithm for music analysis, extracting frames and audio in 30-to-60-second segments.
WhaleLife's "web surfing" plugin uses web search that requires no API key, and external web page content is isolated through a temporary tool contract, so it will not tamper with system instructions or automatically stuff itself into long-term memory.
WhaleLife's model service uniformly manages dialogue, image, speech, vision and other models through the OpenAI-compatible protocol and the Alibaba Bailian interface.
4 more timeline events
All Facts (14)
WhaleLife's Bilibili sub-plugin has already logged in via QR code to a specially prepared account; the AI can browse homepage recommendations, view following updates, search videos, first read the description and subtitles with timestamps, and then watch the video frames in 30-60 second segments; sound recognition and the interaction switch are all off initially and must be turned on manually.
50%UnverifiedThe WhaleLife project is developed based on Node.js 24 and TypeScript, running the TS source code directly, with the frontend using native JS and CSS.
50%UnverifiedWhaleLife's memory architecture removes the short-term memory layer; when the context reaches its limit, the system triggers automatic consolidation, letting the AI write a diary in the first person and choose for itself which memories to keep.
50%UnverifiedWhaleLife's plugin system adopts the CoreList service injection pattern; each plugin runs in a separate subprocess and communicates with the main trunk via RPC, so a plugin crash does not affect the main trunk or other plugins.
50%UnverifiedWhaleLife's autonomy module allows the AI, when no one has sent a message, to get a chance at set intervals and with dynamic probability to judge for itself whether to act, and it can schedule its own next wake-up time, but must avoid the do-not-disturb periods set by the user.
50%UnverifiedWhaleLife's QQ plugin can reuse the QQ already logged in on the computer via the built-in NapCat desktop component, or connect to an external service, and within the whitelisted friends it can chat privately, view images, transcribe voice, and send Bilibili video mini-program cards.
50%UnverifiedWhen saving images, WhaleLife's AI calls a vision model to generate descriptions and index them, storing them in a separate local image library (GESQ Lite), divided into two categories: ordinary images and meme assets.
50%UnverifiedWhaleLife's "audiovisual understanding" skill has FFmpeg built in, uses a quantized SenseVoice plus a VAD model for local speech recognition, and uses a pure TypeScript algorithm for music analysis, extracting frames and audio in 30-to-60-second segments.
50%UnverifiedWhaleLife's "web surfing" plugin uses web search that requires no API key, and external web page content is isolated through a temporary tool contract, so it will not tamper with system instructions or automatically stuff itself into long-term memory.
50%UnverifiedWhaleLife's model service uniformly manages dialogue, image, speech, vision and other models through the OpenAI-compatible protocol and the Alibaba Bailian interface.
50%UnverifiedWhaleLife has made caching optimizations for context; over a chat of more than 100,000 tokens the cache hit rate can reach above 98%, and the cost of a single chat is only a little over three li of a yuan.
50%UnverifiedWhaleLife's fact module has a 100-character limit, but the recorded content is entirely decided by the AI itself rather than being a memo imposed by the user.
50%UnverifiedThe UP host originally planned to open-source WhaleLife only after 50,000 views, but ultimately decided to open-source it early.
50%UnverifiedBoth the main trunk and the demo plugins of the WhaleLife project have been open-sourced under the AGPL-3.0 license in separate repositories on GitHub.
50%