Verbum is a commercial-grade local voice dictation tool tailored for developers, engineers, and AI coding agents (Claude Code, VS Code, Cursor, Windsurf, Windows Terminal, Obsidian). Speak Russian naturally with mixed English technical terms, code identifiers, CLI flags, and German terminology — instantly typed into your active application without accidental execution.
Standard voice typing fails in code editors because it translates code names, garbles technical English, and triggers accidental commands. Verbum solves this with a purpose-built local architecture.
Fast Mode: Sub-300ms single-pass recognition using optimized models for rapid agent prompts.
Accurate Mode: large-v3-turbo-q5_0 with confidence rescue and spoken punctuation for long-form engineering dictation.
Never auto-submits dangerous commands. Verbum strictly suppresses Enter, Tab, and control characters in terminal targets (Windows Terminal, PowerShell, WSL, Claude CLI, Codex).
Speak Russian sentences with mixed English identifiers, camelCase, snake_case, CLI flags (--check), and German terms. Automatic language detection without manual hotkey toggling.
Powered by local whisper.cpp with Vulkan GPU acceleration (AMD Radeon, NVIDIA RTX, Intel Arc) with automatic CPU fallback. Zero audio streaming, zero telemetry, zero mandatory cloud accounts.
Full-format snapshot and restoration for text, rich HTML, DIB bitmaps, and HDROP file paths. Your active clipboard history remains completely untouched during voice injection.
Strict resolution hierarchy: Personal Dictionary > Cryptographically Signed .vpack Archives > Core Glossaries. Custom domain dictionaries for IT, backend, devops, and science.
Why developers replace cloud voice APIs and standard OS voice typing with Verbum.
| Feature / Capability | Verbum v1.2.0 | Cloud Whisper / API | Windows Voice (Win+H) | macOS Dictation |
|---|---|---|---|---|
| 100% Local & Offline | ✓ Yes (Vulkan GPU) | ✗ Cloud Streamed | ✗ Microsoft Cloud | ~ Mixed |
| Zero Layout Switching (RU + EN Code) | ✓ Native Multi-Vocab | ~ Inconsistent | ✗ Single Language | ✗ Single Language |
| Terminal Zero-Execution Guard | ✓ Hardware-level Lock | ✗ No Guard | ✗ May submit Enter | ✗ No Guard |
| Safe Clipboard Snapshot/Restore | ✓ Text + Bitmaps + Files | ✗ Corrupts Clip | ✗ N/A | ✗ N/A |
| Domain Packs (.vpack) & Custom Glossaries | ✓ Signed Packs | ✗ None | ✗ None | ✗ None |
| Latency (Push-to-Talk) | ✓ ~200–400 ms | ✗ 800–2500 ms (RTT) | ~ 500–1200 ms | ~ 400–800 ms |
| Subscription / Per-Minute Cost | ✓ 100% Free / Local | ✗ $0.006/min + API | ✓ Included | ✓ Included |
Verbum is compiled natively for Windows 64-bit architecture with modern hardware acceleration.
Windows 11 (22H2 or newer recommended) or Windows 10 x64 (build 19041+). Native WinUI 3 desktop shell with Mica material support.
Recommended: AMD Radeon (780M / RDNA2 / RDNA3+), NVIDIA GeForce (RTX / GTX 10-series+), Intel Arc with Vulkan 1.3 support.
CPU Fallback: AVX2 / AVX-512 multi-threaded CPU backend included.
RAM: 8 GB minimum (16 GB recommended for Turbo model).
Disk Space: 2.5 GB free space for bundled models and runtime binaries.
Standalone, clean Inno Setup installer. Bundled with complete offline STT models, Vulkan GPU engine, and desktop app.
Download Verbum-Setup-1.2.0.exeGet-FileHash Verbum-Setup-1.2.0.exe -Algorithm SHA256
.vpack files are signed, tamper-evident archives containing domain-specific vocabularies (e.g. Kubernetes, React, Python, Machine Learning). When loaded, the terminology engine canonicalizes spoken phrases (e.g. "докер компоуз" → "docker compose") using an exact word-boundary graph with zero hallucination.