OpenCode: the open-source Claude Code alternative.

◢ The stack

Anthropic API

Claude API for the smartest agent reasoning

$5 free credit · pay-as-you-go

OpenAI

GPT models, embeddings, function calling

Pay-as-you-go

GitHub

Source control + Vercel auto-deploy on push

Free

◢ The build · 5 steps · 18 min

Follow these in order. Don't skip.

Step 01 / 05

Why OpenCode (and when to pick it over Claude Code)

▸Open-source — you can read every line, fork it, run it air-gapped
▸Provider-agnostic — Anthropic, OpenAI, Groq, Together, Ollama (local), all from one config
▸Same TUI vibe as Claude Code — slash commands, plan/execute mode, MCP support
▸Pick OpenCode when: you need on-prem, you want to A/B providers, you want to run a local model on a laptop with no API
▸Stick with Claude Code when: you're shipping fast and don't care about the abstraction layer

Step 02 / 05

Install OpenCode

Terminal

1# macOS / Linux — official installer
2curl -fsSL https://opencode.ai/install | bash
3 
4# Or via npm
5npm install -g opencode-ai
6 
7# Verify
8opencode --version

Step 03 / 05

Configure your provider — Anthropic, OpenAI, or local

~/.config/opencode/opencode.json

1{
2  "$schema": "https://opencode.ai/config.json",
3  "model": "anthropic/claude-sonnet-4-6",
4  "provider": {
5    "anthropic": {
6      "options": { "apiKey": "sk-ant-..." }
7    },
8    "openai": {
9      "options": { "apiKey": "sk-..." }
10    },
11    "ollama": {
12      "options": { "baseURL": "http://localhost:11434/v1" }
13    }
14  },
15  "permissions": {
16    "edit": "ask",
17    "bash": {
18      "git status": "allow",
19      "git diff *": "allow",
20      "npm run *": "allow",
21      "rm -rf *": "deny",
22      "*": "ask"
23    }
24  }
25}

◆ Heads up

Switch models inline: opencode --model openai/gpt-5 — same session, different brain. Useful for A/B testing reliability across providers.

Step 04 / 05

Run a local model with Ollama

The whole point of OpenCode: $0 inference once you're on a local model.

Terminal

1# Install Ollama
2brew install ollama
3 
4# Pull a coding-tuned model
5ollama pull qwen2.5-coder:32b
6 
7# Or NousResearch's Hermes (agentic-tuned)
8ollama pull nous-hermes-2:34b
9 
10# Start the Ollama server
11ollama serve
12 
13# Now point OpenCode at it
14opencode --model "ollama/qwen2.5-coder:32b"

◆ Watch out

Local models are 80% as good as Sonnet on routine code tasks and 30% as good on complex refactors. Use them for boilerplate. Use Claude/GPT-5 for the hard stuff.

Step 05 / 05

AGENTS.md works in OpenCode too

Same convention as Codex CLI — drop an AGENTS.md at your repo root. OpenCode reads it on every session. The format is identical, so a single file works for Codex AND OpenCode.

◆ Ship-it checklist

6 CHECKS

OpenCode installed
At least one provider configured (Anthropic OR OpenAI)
Optional: Ollama + a local model wired in for free inference
Permissions config tuned (read-only auto-allow, destructive deny)
AGENTS.md at your repo root
You ran the same task on two providers and compared outputs

← All guides Show your build in the community →