Portable AI configuration for opencode CLI
Zero API keys · 12 verified models · 27 providers · All tested

opencode · Quick Start · Features · Keyless Models · All Providers · Commands · Skills · Failover · FAQ

---

📑 Table of Contents

1. Quick Start
2. Features
3. Keyless Models
4. All Providers
5. Custom Commands
6. Agent Skills
7. Model Failover
8. FAQ
9. Contributing
10. License

---

Quick Start

curl -sL https://raw.githubusercontent.com/bulbaczPL/opencode-portable/main/setup.sh | bash

Then run:

opencode

Requirements: Linux or WSL, internet connection (required at all times — all models run in the cloud via G4F + Pollinations). Everything else auto-installs.

Tip

Run opencode directly after install — the default model gpt-4o-mini works immediately via local G4F. No API keys needed.

---

Features

🔑 Zero API keys required — 12 models work immediately via G4F + Pollinations 🧠 12 verified text models — each tested live via curl, no guesswork 🌐 27 AI providers — keyless + free-tier with API key ⚡ Auto failover — transparent fallback when a model hits rate limits 🐳 Docker-ready — G4F runs in a container, no dependency hell

🔧 8 custom commands — /status, /switch-model, /g4f-health, more 🧩 3 agent skills — G4F management, keyless providers, registration guides 📦 Single install — curl | bash, auto-installs everything 🎯 OpenCode-native — designed for opencode CLI, uses native config format 🔌 OpenAI-compatible — works with any tool that speaks OpenAI API

---

Keyless Models

These models work right now, with zero configuration, no API key.

Provider	Models	Tested	Notes
G4F (local, Docker)	gpt-4o-mini gpt-4o gpt-4 deepseek-r1 o1 o3-mini command-a command-r command-r-plus command-r7b aria r1-1776	✅ live curl test (31/31 tests)	12 text models — requires G4F running on localhost:1337
Pollinations.ai	GPT-OSS 20B (tool-capable, reasoning)	✅ live curl test	1 concurrent request per IP — text.pollinations.ai

Warning

The following providers were previously listed as keyless but are currently broken (verified July 2026):

• Kilo Gateway — connection timeout
• OVHcloud AI Endpoints — HTTP 404
• ApiAirforce — requires auth (401)
• KeylessAI — Cloudflare 1042 error
• LLM7.io — HTTP 405
• AI Horde — dead endpoint

How to start G4F:

# Option A: Docker (recommended)
docker run -d --name g4f --rm -p 1337:1337 python:3.12-slim sh -c \
  "pip install -q g4f==7.7.2 nest-asyncio2 && \
   python3 -c 'from g4f.api import run_api; run_api(port=1337, host=\"0.0.0.0\")'"

# Option B: Direct Python
python3 -c "from g4f.api import run_api; run_api(port=1337)"

---

All Providers

Legend

Color	Meaning
🟢	Keyless — no API key needed
🟡	Free tier — API key required, no credit card
🔴	Broken — kept for reference

Provider table

Provider	Tier	Models	Key needed
G4F (local)	🟢	12 verified text models (31/31 tests ✅)	—
Pollinations.ai	🟢	GPT-OSS 20B	—
Groq	🟡	Llama 3.3, Gemma 2, Mixtral, DeepSeek-R1-distill	Free key
Cerebras	🟡	Llama, Qwen (fastest inference)	Free key
Mistral	🟡	Mistral Small, Codestral	Free key
NVIDIA NIM	🟡	100+ models, Nemotron 3, Llama	Free key
OpenRouter	🟡	~22 free models (:free suffix)	Free key
Cloudflare Workers AI	🟡	50+ models, Kimi K2.6, Llama, GPT-OSS	Free key
Together AI	🟡	Llama 3.3, Qwen, Gemma 4, GPT-OSS	Free key
SambaNova	🟡	DeepSeek V3.1, Llama 3.3, GPT-OSS	Free key
Hugging Face	🟡	DeepSeek V3, Llama 3.3, Phi-3.5	Free key
GitHub Models	🟡	GPT-5, GPT-4.1, o4-mini, Llama 4	Free key
DeepSeek	🟡	DeepSeek V4 Flash, V4 Pro	Free key
Scaleway	🟡	Llama 3.3, Qwen Coder, Mistral, DeepSeek R1	Free key
Cohere	🟡	Command A+, Command A, Command R+	Free trial
SiliconFlow	🟡	Qwen 3 8B, DeepSeek R1 Distill	Free key
Z.AI (Zhipu)	🟡	GLM 4.7 Flash, GLM 4.6V Flash	Free key
ModelScope	🟡	Qwen 3.5 35B, Qwen 3.5 27B	Free key
DashScope	🟡	Qwen 3.6 27B, Qwen 2.5 Coder 32B	Free key
Ollama Cloud	🟡	GPT-OSS 120B, DeepSeek V3.1, Qwen3 Coder	Free key
FreeTheAi	🟡	GPT-OSS 20B, Llama 3.3, Qwen3 Coder	Discord key
Aion Labs	🟡	Aion 2.5, Aion-RP 1.0	Free key
Token-Free Gateway	🟡	Claude, ChatGPT, Gemini, DeepSeek, Grok	Browser login
Kilo Gateway	🔴	—	—
OVHcloud	🔴	—	—
ApiAirforce	🔴	—	—
smanx-free	🔴	—	—

---

Custom Commands

Available in opencode TUI. Type /help to see all commands.

Command	Description
/status	Show current model, provider status, API key status
/switch-model <nr|name>	Switch to a specific model or provider
/g4f-start	Start G4F aggregator (auto-detect Docker or Python)
/g4f-health	G4F health check, restart, view logs
/fallback-chain	View or configure the failover chain
/fallback-chain test	Test all providers in the chain
/provider-test	Test all 27 providers
/provider-register	Step-by-step guide for free-tier provider registration
/auto-failover on|off	Toggle auto-failover

---

Agent Skills

Skills are automatically available in opencode. They provide specialized know-how for agent operations.

Skill	Description
g4f-management	Start, stop, restart, troubleshoot G4F aggregator
keyless-providers	Rate limits, reliability data for all keyless providers
provider-registration	Step-by-step registration guides for all free-tier providers

---

Model Failover

When the current model fails (rate limit, timeout, auth error), opencode-portable automatically falls through this chain:

Keyless (G4F → Pollinations)
  ↓
OpenRouter free
  ↓
Free-tier API keys (Groq → Cerebras → Mistral → NVIDIA → …)
  ↓
Specialized (Cohere → SiliconFlow → Z.AI → …)

The router tries each provider in order. On success it stops; on failure it moves to the next. Configure via /fallback-chain in TUI.

---

FAQ

Why only 2 keyless providers?

We tested over 20 claimed "keyless" endpoints from Reddit, GitHub, and Discord. Only G4F and Pollinations.ai returned valid responses without any form of authentication. The rest required API keys, returned empty responses, or were dead.

Can I still use the broken providers?

They remain in the config for reference. If their endpoints change, open them. Do not rely on them.

Which G4F models were tested and failed?

Of the 24 originally listed G4F models, only 6 passed live testing. The remaining 18 passed — the others (claude-, gemini-, llama-*, grok-3, qwen-2.5-coder-32b, etc.) all returned HTTP 500/401/404 or empty responses.

How do I add API keys?

In opencode TUI, run /connect. Or manually add to ~/.config/opencode/opencode.jsonc:

"providers": {
  "groq": {
    "id": "@ai-sdk/openai-compatible",
    "apiKey": "${env:GROQ_API_KEY}",
    "options": {
      "baseURL": "https://api.groq.com/openai/v1"
    }
  }
}

Each provider's key signup page is listed in the All Providers table.

Does this work on macOS or Windows?

Linux and WSL only. You can run opencode-portable in Docker on any platform.

---

Contributing

Contributions are welcome! Found a working keyless endpoint? Opened an issue or PR. Found a broken model? Same.

• Open an issue
• Submit a PR

---

License

MIT © 2026 bulbaczPL

---

---

🇵🇱 Polski — przeczytaj ten dokument po polsku

Tłumaczenie w przygotowaniu. W międzyczasie możesz przetłumaczyć stronę przez przeglądarkę (Ctrl+klik → Przetłumacz na polski).

🇩🇪 Deutsch — dieses Dokument auf Deutsch lesen

Übersetzung in Vorbereitung. Du kannst die Seite im Browser übersetzen lassen.

🇫🇷 Français — lire ce document en français

Traduction en préparation. Vous pouvez traduire la page via le navigateur.

🇪🇸 Español — leer este documento en español

Traducción en preparación. Puedes traducir la página con el navegador.

🇮🇹 Italiano — leggere questo documento in italiano

Traduzione in preparazione. Puoi tradurre la pagina tramite il browser.

🇵🇹 Português — ler este documento em português

Tradução em preparação. Podes traduzir a página através do navegador.

🇷🇺 Русский — прочитать этот документ на русском

Перевод готовится. Вы можете перевести страницу в браузере.

🇨🇳 中文 — 阅读此文档的中文版本

翻译准备中。您可以通过浏览器翻译此页面。

🇯🇵 日本語 — 日本語でこのドキュメントを読む

翻訳準備中。ブラウザでページを翻訳してください。