August 31, 2026
v0.7.5- Three new model choices, all of which can read images: DeepSeek V4 Flash at roughly a fifth of the cost per token, GLM 5.3 Flash with over a million tokens of context, and Qwen3.8 27B with 262k tokens of context.
- Leo can look at images you upload on every workspace, not just some of them.
- When you attach an image, Leo picks the right model for your workspace instead of always switching to the same one.
- New models can be added from a settings file, so trying one no longer waits on a software update.
- Leo's answer survives a dropped connection instead of cutting off mid-sentence.
- A model that stops responding is dropped in 90 seconds instead of 10 minutes, and Leo finishes the answer on a second model.
- Leo tells you when it is retrying, instead of appearing to sit still.
- Your app restarts itself after a crash instead of staying down.
- When Leo asks several setup questions at once, you answer them one at a time with Back and Next.
- Cookbook search now includes the recipes you saved yourself, not only the public ones.
- The feedback button keeps what you typed and retries if your session has timed out.
- Files uploaded to the wrong folder survive an update instead of disappearing.
- Fixed a sign-in failure that could happen depending on how your browser makes its first request.