AI coding tools evolve fast, and OpenCode stands out as an open-source agent rivaling Cursor or Claude Code CLI - fully terminal-native with LSP support, multi-session handling, and 75+ LLM providers via Models.dev.

Compared to locked-in alternatives that are subscription base you can pick models pay-as-you-go prioritizing privacy and flexibility.
Check my 5-minute setup video for a quick demo.

https://youtu.be/VSN3fCJcoIc

Installation

Start with the one-liner curl install - no Docker or complex deps needed:

curl -fsSL https://opencode.ai/install | bash ​.

This adds the opencode CLI globally. Verify with opencode --version, then launch via opencode. For VS Code integration, it pairs seamlessly as an LSP client. Full docs: opencode.ai/docs.​​

OpenRouter Setup

OpenRouter powers model access via its OpenAI-compatible API - sign up at openrouter.ai, grab a key, and fund credits (pay-per-token, starting free tiers).​​​
Export these env vars (add to ~/.zshrc for persistence):

export OPENAI_API_BASE=https://openrouter.ai/api/v1
export OPENAI_API_KEY=your-openrouter-api-key

Source your shell (source ~/.zshrc) and relaunch. Screenshot your API keys page for readers to match. This unlocks 300+ models without switching providers.
I added some credits to OpenRouter just for more advanced models for different projects.

Source your shell (source ~/.zshrc) and relaunch. Screenshot your API keys page for readers to match. This unlocks 300+ models without switching providers.

Basic Usage

Run opencode in a repo to init: It prompts project context, then use slash commands in the TUI.​

• /connect: Link OpenRouter (auto-detects env vars).
• /models: List/select (e.g., Claude 3.5 Sonnet or GPT-4o-filter free ones).
• /init: Scans repo, suggests plan (e.g., "Analyze main.rs and create layers").
• Core flow: /plan for outline, /build to generate/run code, /improve for refinements.