Saymore turns speech into text at the current cursor, so you can dictate in the
editor, browser, chat app, terminal, or other text field you already use. Press
a global shortcut, speak naturally, and Saymore recognizes, optionally refines,
and inserts the result without opening a separate writing surface.
It is a native Rust and Slint desktop application with
explicit boundaries between speech recognition, text refinement, storage, and
platform integration. You choose the providers; Saymore does not require a
hosted Saymore account or backend.
Project status: Saymore is usable and distributed for macOS and Windows,
but remains under active development. The latest release may lag behind the
repository state.
What You Can Do
| Area | Available today |
|---|
| Cross-app dictation | Start and stop from a configurable global shortcut, then insert text into the focused editable control. Defaults are Right Command on macOS and Right Alt on Windows. |
| Speech recognition | Use macOS Dictation, Volcengine, or a custom OpenAI-compatible speech endpoint. Availability depends on the operating system and provider configuration. |
| Optional refinement | Conservatively clean up filler, punctuation, self-corrections, and structure with SenseNova or DeepSeek. Invalid or unavailable refinement falls back to the recognized text. |
| Personal dictionary | Add, edit, delete, search, filter, and import canonical spellings from CSV. Saymore can also learn terms from repeated corrections in supported text controls. |
| Private history | Search, inspect, copy, delete, or clear encrypted local history, with configurable retention. Original audio is not stored. |
| Desktop controls | Onboarding, permission checks, microphone selection, multiple shortcuts, launch at login, system tray controls, themes, English and Simplified Chinese UI, update checks, and privacy-safe diagnostic export. |
When text insertion cannot be verified, Saymore keeps the final transcript
available for recovery instead of silently discarding it.
How It Works
Global shortcut
-> record speech in memory
-> recognize with the configured ASR provider
-> apply deterministic local cleanup
-> optionally refine with the configured LLM provider
-> normalize confirmed dictionary spellings
-> insert at the current cursor
The refinement stage is deliberately narrow: it improves a transcript without
answering questions, inventing facts, or turning dictation into a chatbot. If a
provider fails or its output violates Saymore's safeguards, the pipeline falls
back to the last safe text.
Privacy Model
- Audio is held in memory for recognition and is not written to local history.
- Cloud ASR sends audio to the provider you configure. Cloud refinement sends
the transcript and only relevant confirmed dictionary terms to the selected
LLM provider.
- Local history is encrypted. Its encryption key is kept in the platform
credential store, and retention can be disabled or changed by the user.
- Sensitive controls are treated specially and are excluded from history and
correction learning.
- Diagnostics stay local and record allowlisted event identifiers, not
transcripts, API keys, device names, paths, or raw error details.
- Saymore does not read screen context, compose replies, or automatically send
messages.
The product direction
defines the complete data, provider, and feature boundaries.
Download
| Platform | Package |
|---|
| macOS 12+ | A signed and notarized universal DMG for Apple Silicon and Intel Macs is available from GitHub Releases. |
| Windows | A current-user NSIS installer is available from GitHub Releases. |
Each direct release includes SHA256SUMS for artifact verification. A Mac App
Store release is planned, but is not available yet.
After installation:
- Complete onboarding and grant microphone permission. macOS also requires
Accessibility permission for global shortcuts and cross-app text insertion.
- Choose and test a speech-recognition provider on the Models page.
- Focus any editable field and use the configured shortcut to start and stop
dictation. Press Escape to cancel an active recording.
Development
The production application uses Rust and Slint. Node.js, a WebView, and a web
frontend are not part of the build.
On macOS, start the persistent signed development preview:
./scripts/dev-preview.sh
On Windows, build the desktop application with Cargo:
cargo build -p saymore-desktop
The workspace has four ownership boundaries:
| Path | Responsibility |
|---|
crates/app | Business types, invariants, use cases, and port traits |
crates/infra | Filesystem, database, network, audio, and platform adapters |
apps/desktop | Slint UI, dependency wiring, and process lifecycle |
crates/xtask | Repository maintenance and packaging automation |
desktop -> app
desktop -> infra -> app
See the development guide for prerequisites, preview and
packaging workflows, and the complete quality gate. See
Architecture for crate ownership and platform
boundaries.
Documentation
The documentation index links to the complete set of product,
engineering, ADR, and research documents.
Contributing
Issues, design discussions, documentation feedback, and reproducible bug reports
are welcome. Please read CONTRIBUTING.md before starting
implementation work. The external Contributor License Agreement workflow is not
yet available, so code contributions currently require prior coordination with
the maintainers.
License
Saymore is source-available, not OSI-approved open source. It is licensed
under the PolyForm Shield License 1.0.0. Personal, internal
organizational, and other noncompeting uses are permitted. Providing a product
or service that competes with Saymore requires a separate commercial license
from the maintainers. Third-party assets retain their own licenses.
