OpenCode Config

My personalized OpenCode configuration.

Files

• opencode.json — Main OpenCode configuration file. Defines MCP servers, permissions, and settings.
• AGENTS.md — Global agent instructions (e.g., Context7 MCP usage guidelines).
• commands/review-pr.md — Custom review-pr command for handling PR checks and review feedback.
• package.json — Dependencies required by this config (e.g., @opencode-ai/plugin, @slkiser/opencode-quota).
• plugins/opencode-voice/ — Vendored @renjfk/opencode-voice with Linux audio fixes applied.
• tui.json — TUI-specific configuration (e.g., voice plugin).

Setup

1. Clone this repo to your machine.
2. Symlink or copy the files into your OpenCode config directory (usually ~/.config/opencode/):

   ln -s $(pwd)/opencode.json ~/.config/opencode/opencode.json
   ln -s $(pwd)/AGENTS.md ~/.config/opencode/AGENTS.md
   ln -s $(pwd)/commands ~/.config/opencode/commands

3. Update opencode.json to replace machine-specific paths with paths valid on your system.

Note: The excalidraw MCP server in opencode.json contains absolute paths specific to the original machine. Update the command array paths (node binary and mcp_excalidraw script) to match your environment.

Plugins

@slkiser/opencode-quota

Quota, usage, and token visibility for OpenCode with zero context window pollution.

• Shows popup quota toasts after assistant responses
• Adds TUI sidebar panel with quota rows
• Provides /quota, /quota_status, and /tokens_* commands
• Supports providers: GitHub Copilot, OpenAI, Cursor, Anthropic, and more

After syncing this config and installing dependencies, restart OpenCode and run /quota_status to verify.

@renjfk/opencode-voice (vendored with Linux fixes)

Speech-to-text and text-to-speech plugin for OpenCode. This repo includes a vendored copy under plugins/opencode-voice/ with patches for Linux audio device enumeration and recording (the upstream only supports macOS CoreAudio).

• STT: Record voice prompts with local Whisper transcription, normalized by an LLM
• TTS: Hear assistant responses spoken aloud via Piper TTS
• Keybinds: ctrl+r to record, ctrl+x then s to speak last response, ctrl+x then v to toggle auto TTS

Voice Prerequisites (Linux)

The plugin requires system-level binaries that are not installed by npm:

1. sox — for audio recording and playback
2. whisper-cli — for local speech-to-text transcription
3. piper — for local text-to-speech synthesis
4. Voice models — Whisper and Piper voice model files

A convenience script is provided to automate this on Debian/Ubuntu:

./scripts/setup-voice.sh

Note: The script builds whisper-cli from source and downloads models (~650 MB total). It also requires uv (or pipx) for installing Piper.

Voice Configuration

The tui.json in this repo configures the voice plugin with Anthropic as the normalization LLM. Ensure ANTHROPIC_API_KEY is set in your environment. You can change the endpoint/model in tui.json to any OpenAI-compatible API (Ollama, vLLM, etc.).

Security

This repository is configured to never commit sensitive data:

• .env files are ignored.
• No API keys, tokens, or credentials are stored in tracked files.
• Environment variables in opencode.json are limited to non-sensitive values (e.g., EXPRESS_SERVER_URL=http://localhost:3000).

If you add MCP servers that require secrets, use environment variables or a local .env file (which is gitignored) rather than hardcoding values.