Changelog - esper

December 1, 2025

Major code refactor and optimization. This allowed for some optimizations and fixes like: major initialization speed improvements for both online and offline models, streaming fixes with some providers, and more.
Introducing esper Pro! A new one time purchase allows you to download more than one offline model at a time and also gives access to theme customization and more. Anyone who has already purchased the app or downloaded before December 1st will automatically be upgraded to Pro for free. This will allow us to continue to support the app and add new features in the future!
Added new GPU Acceleration option for offline models. This allows you to choose between faster GPU inference and more battery efficient CPU inference.
Added new toggle for resetting advanced parameters between chats. This allows you to turn it off so that params are saved everytime you edit them and then applied to any new chats you create.
Moved these two new settings and other preferences into their own settings menu screen.
Updated licenses page.

Fixed an issue causing the app to overload RAM when downloading large models, sometimes causing it to crash.
Fixed several error message issues and improved error handling.

November 21, 2025

November 13, 2025

Voice mode 🎤. Talk with any model using your voice.
Updated Google Gemini API support. More thinking controls on relevant models, more file upload options, Google web and Google maps search grounding, automatic url retrieval.
Updated xAI Grok search options. Select between searching the web or X (or both).
New models: Qwen3 VL vision model variants and IBM Granite text model.
Better conversation search. Search now keyword matches over messages as well as chat titles.

October 13, 2025

Added Apple Foundation model support.
Added TTS voice selection.
Added local model verify button. This feature makes sure the local model and filesystem are in sync.

September 18, 2025

Major framework and package updates.
iOS 26 upgrade support.
Complete refactor of the underlying chat UI, increasing performance and stability.
Added image generation support for OpenAI, xAI, OpenRouter, and Google API providers.

Fixed an issue where the interface would be very slow with large conversations.
Fixed accessibility issues with the switches in the Share With AI screen.
Fixed an issue causing Liquid AI LFM2-VL models to not work.
Fixed an issue causing the app to slow when deleting or downloading local models.
Fixed an issue with long system prompts not being editable in the My GPTs screen.
Fixed an issue causing assistant messages to copy plain text instead of full markdown.
Fixed an issue causing message generations intentionally stopped by the user to not be saved.