Back home

OpenCode Hex

x 0.85
Visit source

Intro

Hex (often called OpenCode Hex) is a free, open-source, local-first voice dictation app from Anomaly, the company behind the OpenCode coding agent. Its first stable cross-platform release, Hex 3.0.0, shipped on October 31, 2026, and the official @opencode account announced it the same day — the post drew more than 4,200 likes and 240,000 views.

The idea is simple: hold a shortcut (Option on Mac), say what you mean, release, and Hex transcribes your speech on your own device and pastes the text wherever your cursor is — a terminal, an editor, a chat box or an AI agent prompt. It is written in native Rust, runs on Apple-silicon macOS 15+, Windows (x64/ARM64) and Linux (x64/ARM64), and is MIT-licensed. Many users immediately framed it as a free alternative to paid dictation apps such as Wispr Flow.

Explain

What's in Hex 3.0

  • On-device transcription: download a local speech model during setup (e.g. Parakeet, Whisper, and new Underdog Bark 3.0 Pro / Clean models); ordinary dictation needs no cloud service and no OpenCode install. Local models now use about 160 MB of memory instead of ~900 MB.
  • Streaming on Windows/Linux: with a Vulkan GPU driver, Parakeet streams while you speak, so text arrives ~100 ms after you release the key.
  • Unified settings on all three OSes: General, Output (replacements, per-app modes, OpenCode rewriting, transformations) and History.
  • Vocabulary: list names and jargon so Hex steers transcription toward them and fixes common mishearings.
  • Flexible shortcuts: both Option keys, a left/right modifier, or a mouse button; Hold or Toggle modes.
  • hex CLI: hex status, hex dictation, hex history, hex meetings, and an hex mcp server so agents can read your dictation history.
  • Meetings (beta, macOS Labs): record calls, calendar reminders and exported meeting notes.
  • Annotate (Dev channel): record the screen while you draw and talk, then send it to an OpenCode session.

Optional OpenCode integration

If you have OpenCode V2 installed, Hex can send the transcript (plus the foreground app name) to your configured model to rewrite or clean it up before pasting. If rewriting fails, Hex falls back to the local transcript. This step is opt-in; plain dictation stays fully local.

Upgrade note

Hex 3.0 is effectively a new app (anomaly.hex, signed by Anomaly): it does not migrate settings, history or models from 2.3.1, and on macOS you must re-grant Microphone and Accessibility permissions.

Usage

Install

  • macOS (Apple silicon, macOS 15+): download the DMG from https://anoma.ly/hex or run brew install --cask anomalyco/tap/hex.
  • Windows 10/11 (x64/ARM64): download the installer from the website (no admin rights needed).
  • Linux (x64/ARM64): curl -fsSL https://anoma.ly/hex/install-linux.sh | sh, or use the .deb / .rpm packages, Nix, or yay -S hex-bin on Arch.

First run

  1. Grant Microphone and Accessibility permissions (macOS).
  2. Pick a language and a local speech model; the model downloads once.
  3. Choose your shortcut (default: hold Option on Mac, Ctrl+Space on Windows).
  4. Hold the shortcut, speak, release — the text is pasted at your cursor.

Tips

  • Add product names, teammates' names and code terms to Vocabulary for better accuracy.
  • Use per-app Modes, e.g. keep raw text in the terminal but enable OpenCode rewriting in Slack or email.
  • Pair it with coding agents (OpenCode, Claude Code, Codex): speaking a long task description is often faster than typing it.
  • Use hex mcp to let an agent search what you dictated earlier.

Links