Back to all entries
July 10, 2026
The biggest update yet: connect any AI provider (including local and on-device), a reworked settings experience, a movable recording bar, and a dramatically more capable API.
New: Providers
- New Providers tab: connect providers from a clean grid — click, paste your key, it validates, done.
- Ollama and LM Studio support: run AI features against local models. Sotto auto-detects running instances and lists your installed models.
- Apple Intelligence: auto-detected on supported Macs. AI features work out of the box — no API key needed.
- New transcription providers: Deepgram Nova-3 and custom OpenAI-compatible endpoints (self-hosted Whisper, Speaches, LiteLLM…).
- All model catalogs updated to the latest generation (GPT‑5.6 family, Claude Sonnet 5 / Opus 4.8 / Fable 5, Gemini 3.5, and more). Your saved models migrate automatically.
- Use your coding agent CLIs as providers: Codex, Claude Code, OpenCode, Grok, and Cursor Agent.
New: Permissions
- New Permissions tab showing every permission Sotto uses, with one-click granting.
- A floating guide helps you drag Sotto straight into the right System Settings list — no more hunting for the app.
- "Repair Permission" fixes stale Accessibility grants.
Recording bar
- Drag it anywhere — Sotto remembers where you put it.
- Right-click anywhere on the bar for a full menu: size, style, position, reset.
- "Hide for This Recording" records invisibly, just this once.
Fixes
- Global hotkey now works immediately after login, before the main window is ever opened.
- Fixed recordings getting stuck ("bubble stays on screen") until the app was restarted.
- Errors are shown instead of failing silently.
- License activation shows real error messages, and deactivation tells you if the seat wasn't freed.
API & CLI (for tinkerers and agents)
- Browse and search your transcription history via the API.
- Manage AI functions, dictionary, find & replace, and always-on rules via API and CLI.
- Upload audio files directly (
POST /api/import/upload) — works remotely and around sandbox limits. - Endpoints grouped with per-group and per-endpoint enable/disable.
- The
sottoCLI auto-discovers the port, supports remote hosts (--host), and gained commands for everything above.
Settings got a cleaner grouped sidebar, Summaries and Titles are now separate (and titles work for dictation recordings too), and AI Functions include starter templates.