Providers & Models

agentty ships with Claude as the default, but it speaks to any OpenAI-compatible backend. Pick one with --provider, or switch live mid-thread with ^P (provider) and ^/ (model).

The default: Claude

With no flags, agentty talks to Anthropic using your Claude Pro/Max OAuth subscription or an sk-ant-… API key. Nothing to configure.

agentty                        # Claude, OAuth or API key
agentty -m claude-opus-4-5     # pick a specific Claude model⧉

Switching providers

--provider selects the backend. It is persisted between runs just like -m, so you only pass it when you want to change it.

agentty --provider openai -m gpt-4o        # GPT
agentty --provider groq -m llama-3.3-70b   # Groq
agentty --provider ollama -m qwen2.5-coder # local model, no key
agentty --provider openrouter              # any model via OpenRouter⧉

Inside a thread, press ^P to switch provider and ^/ to switch model — no restart, no re-auth. Both are also reachable from the command palette (^K). The next turn uses the new backend.

Supported providers

API keys

Hosted OpenAI-compatible providers read their key from the provider-specific environment variable (e.g. GROQ_API_KEY), falling back to OPENAI_API_KEY, or an explicit -k <key> for the session. Ollama needs no key.

export GROQ_API_KEY=gsk_…
agentty --provider groq -m llama-3.3-70b

# or a one-off, never written to disk:
agentty --provider openai -k sk-… -m gpt-4o⧉

Local models (Ollama)

Point agentty at a model served by Ollama on localhost:11434 — no key, no cloud, no data leaving your machine. agentty uses Ollama's native /api/chat protocol and salvages tool calls that weaker local models leak as raw JSON, so even smaller models can drive the full tool suite.

ollama pull qwen2.5-coder
agentty --provider ollama -m qwen2.5-coder⧉