Local Waifu

Local Waifu is a desktop AI companion for macOS and Windows that runs on your own hardware. The language model, long-term memory, image generation, voice input and speech can all run locally. There is no account to create and no telemetry. Even if you block every outbound connection, the app keeps working.

The app includes its own local model runtime, so there is no separate setup required. On first launch, it detects your hardware, recommends a model that fits the available memory, and downloads it for you — roughly 3–22 GB depending on the model tier.

If you prefer hosted models, an optional bring-your-own-key mode supports OpenAI, Anthropic, Google, DeepSeek, Mistral, Groq, OpenRouter and OpenAI-compatible endpoints. This is completely optional and disabled by default; requests are sent directly to the provider you choose using your own API key.

Memory is at the core of Local Waifu. Conversations are stored locally in SQLite, combined with vector embeddings and an automatically built knowledge graph of the people, places and details you mention. Important information is distilled into a persistent memory that the character can carry into future conversations.

That memory survives app restarts and model changes, so your companion can still remember important details months later instead of starting from scratch every time you open a new chat.

Characters are created rather than chosen from a fixed catalog. You can customize their name, appearance, backstory and personality using eight personality dimensions, including traits such as shy ? flirty, calm ? energetic and reserved ? expressive.

Characters can be waifu, husbando or non-binary. Relationships progress through seven stages based on earned XP, mood is represented by a live six-dimensional state, and characters can shift between 13 conversational modes depending on the situation. You can create multiple characters, each with its own memory and separate conversation history.

Your companion can also see and create images. Selfies are generated locally on your GPU using a bundled Stable Diffusion engine, while images shared in chat can be understood by a vision-capable model.

Voice conversations can run locally as well, including speech recognition and spoken replies. Optional cloud speech providers are available for users who prefer them.

The interface and AI responses support seven languages: English, Polish, German, Spanish, Japanese, Korean and Chinese.

Characters can be exported and moved between computers, so they are not tied to an online account or service.

Local Waifu is sold as a one-time purchase with a 7-day free trial and a 30-day refund period. There is no subscription and, when using local models, no per-message or per-image fees.