4.1.0

4.1.0 is currently a beta release and must be installed manually — the app will not auto-update to it. Download:

macOS (Apple silicon): https://cdn.feelfish.com/feelfish-releases/v4.1.0-beta.2-20261015T135451/FeelFish-arm64.dmg

macOS (Intel): https://cdn.feelfish.com/feelfish-releases/v4.1.0-beta.2-20261015T135451/FeelFish-x64.dmg

Windows (x64): https://cdn.feelfish.com/feelfish-releases/v4.1.0-beta.2-20261015T135451/FeelFish-Setup-4.1.0-beta.2.exe

🎙️ Voice input: hold the space bar to speak, release to transcribe, press Enter to send — recognition runs on your own machine (member feature; the speech model downloads on first use)
🔊 A read-aloud bar with two voices (local model and Doubao) and click-to-locate for the paragraph being read; agent replies can be read aloud too (every reply or only the final round)
🎧 New "Audiobook production" solution: prepare the narration script, cast voices, estimate cost, then batch generate and deliver chapter by chapter
🧰 A new "Media models" section in the Treasure Box: image generation, Doubao speech synthesis and Doubao audio generation, saved to your project automatically, with a configurable API base URL
📊 New media job board: progress and per-step detail with cancel, retry, resume and export; unfinished items resume after a restart, and completed ones are never generated twice
🧩 Custom media APIs (image / video / speech) you can plug in, with a connectivity test and a confirmation before every run
🔌 Custom model APIs support the native Gemini protocol
✍️ Writing and settings refinements: CSV files support table editing (add rows and columns, edit cells directly), credit details show expiry times, cloud sync reports file types it does not overwrite in one notice, and local models can be downloaded, updated, cleaned up or reinstalled
🐛 Fixed long sub-agent sessions failing permanently after context compaction, the HarmonyOS PC soft keyboard covering the editor, and the read-aloud bar occasionally getting stuck on "preparing"
🔧 Other minor detail optimizations