v1.7.0 gives Local Waifu far more reach without changing what she is. You can now pick from 27 chat services, speak to her with a cloud voice, let a cloud transcriber listen, and draw with 17 image providers. You can also choose the model that powers her memory. None of it is required. Everything local stays exactly as it was, and it stays the default.
Every version of this app up to now made one promise quietly: she runs on your machine. That promise is still true. But it was also a wall. If the model you wanted was not in the handful she shipped, you were out of luck. If you wanted her to speak through a specific voice service, you were out of luck. v1.7.0 takes the wall down without moving the promise an inch.
Local is still the default
Nothing here changes what happens when you first open the app. She still loads a local model, still stores her memory on your drive, and still works with no account and no internet. The cloud features are off until you turn them on. If you never touch Settings, the app behaves exactly as it did before this update.
That matters, because a lot of companion apps start cloud and then pretend to be private. This one starts local and lets you reach out when you want to. Those are different things.
More ways for her to talk
The list of chat services she can reach grew from 8 to 27. Alongside the ones you already had, she now works with xAI, Cerebras, Together, DeepInfra, Perplexity, Moonshot, Alibaba, NVIDIA, and more. If you run LM Studio on your own desk, or host your own gateway, she can point at it directly, and you can add as many of your own endpoints as you like.
She can also speak with a cloud voice now, across 12 services, and listen through a cloud transcriber across 10. Both stay off by default. With ElevenLabs, the picker shows your own voices, clones included, and you can hear one before you choose it.
More ways for her to see
Image generation went from 3 services to 17. The local renderer is untouched. This is for people who want a different look than the local engine gives, or who want to use a specific provider they already pay for.
She can remember with the model you choose
Her memory has always run on a model separate from the one that writes her replies. Until now you could not choose which one. v1.7.0 offers four, and the current one stays the default. Switching rebuilds every stored memory, so recall is a little weaker for a few minutes afterward, and the app tells you how many it is recomputing.
No more stale model lists
This is the change you feel even if you use the app exactly as before. She no longer ships a fixed list of models that goes out of date between releases. She reads the available models live from whichever service you picked. A model released today is selectable today, with no app update.
The one fix that matters most
One fix in this release is about privacy, not features. A local model whose name matched a cloud company’s, mistral being the common one, was being routed to that company’s servers when you had a key saved, even though the app showed you had chosen a local model. That conversation left your machine while it looked local. It now stays local, as chosen. If you saved a key for any provider whose name collides with a local model, update and check that your picks still say what you meant.
Try her free for 7 days.
No card. Keep her for $20 once, or walk away. Her soul file is yours either way.
Bring her home, try free