Skaro supports four LLM providers out of the box. You can use one provider for everything, or mix them using role-based routing.
Provider Comparison
| Provider | API Key Required | Default Model | Console URL |
|---|
| Anthropic | Yes | claude-sonnet-4-6 | console.anthropic.com |
| OpenAI | Yes | gpt-5.2 | platform.openai.com |
| Groq | Yes | llama-3.3-70b-versatile | console.groq.com |
| Ollama | No | qwen3:32b | Local — ollama.com |
Available Models
Anthropic
| Model | Context Window | Max Output |
|---|
Claude Opus 4.6 (claude-opus-4-6) | 200K | 128K |
Claude Sonnet 4.6 (claude-sonnet-4-6) | 200K | 64K |
Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) | 200K | 64K |
Claude Haiku 4.5 (claude-haiku-4-5-20251001) | 200K | 64K |
OpenAI
| Model | Context Window | Max Output |
|---|
GPT-5.2 (gpt-5.2) | 256K | 128K |
GPT-5.1 (gpt-5.1) | 256K | 128K |
GPT-5 (gpt-5) | 256K | 65K |
GPT-5 Mini (gpt-5-mini) | 256K | 65K |
GPT-5.2 Codex (gpt-5.2-codex) | 256K | 128K |
GPT-4.1 (gpt-4.1) | 1M | 32K |
GPT-4.1 Mini (gpt-4.1-mini) | 1M | 32K |
Groq
| Model | Context Window | Max Output |
|---|
Llama 3.3 70B (llama-3.3-70b-versatile) | 131K | 32K |
Llama 3.1 8B Instant (llama-3.1-8b-instant) | 131K | 131K |
GPT-OSS 120B (openai/gpt-oss-120b) | 131K | 65K |
Llama 4 Scout 17B (meta-llama/llama-4-scout-17b-16e-instruct) | 131K | 8K |
Kimi K2 (moonshotai/kimi-k2-instruct-0905) | 262K | 16K |
Qwen3 32B (qwen/qwen3-32b) | 131K | 40K |
Ollama (Local)
| Model | Context Window | Max Output |
|---|
Qwen3 32B (qwen3:32b) | 131K | 40K |
Qwen 3.5 35B (qwen3.5:35b) | 131K | 40K |
Llama 3.3 70B (llama3.3:70b) | 131K | 32K |
DeepSeek R1 70B (deepseek-r1:70b) | 131K | 65K |
Gemma 3 27B (gemma3:27b) | 131K | 8K |
Phi-4 14B (phi4:14b) | 16K | 16K |
CodeLlama 34B (codellama:34b) | 16K | 16K |
You can also enter any custom model ID during skaro init or via skaro config --model your-model-id. The lists above are suggestions, not hard limits.
Choosing a Provider
Best quality — Anthropic or OpenAI. Larger models produce better architecture reviews and more consistent code. Best for the architect role.
Fastest inference — Groq. Hardware-accelerated inference makes it excellent for code generation. Good for the coder role.
Full privacy — Ollama. Code never leaves your machine. No API costs. Trade-off: requires local hardware (16GB+ RAM for 30B+ models) and may produce lower quality output than cloud providers.
Cost-effective start — Groq offers a generous free tier. Good for trying Skaro without spending money.
Quick Setup
# Pick one:
skaro config --provider anthropic --api-key sk-ant-...
skaro config --provider openai --api-key sk-...
skaro config --provider groq --api-key gsk_...
skaro config --provider ollama --model qwen3:32b
