Provider setup

LLMSwap 5.7.0 implements 10 provider adapters. You need only one configured provider to start.

Provider	Default model	Environment
Anthropic	claude-sonnet-5	ANTHROPIC_API_KEY
OpenAI	gpt-5.6	OPENAI_API_KEY
Gemini	gemini-3.6-flash	GEMINI_API_KEY
Cohere	command-a-plus-05-2026	COHERE_API_KEY
Perplexity	sonar-pro	PERPLEXITY_API_KEY
IBM watsonx	ibm/granite-4-h-small	WATSONX_API_KEY, WATSONX_PROJECT_ID
Groq	openai/gpt-oss-120b	GROQ_API_KEY
Ollama	qwen3.5:9b	Local server and installed model
xAI	grok-4.5	XAI_API_KEY
Sarvam	sarvam-105b	SARVAM_API_KEY

Defaults were audited on August 3, 2026. Provider lifecycle, account access, regional availability, and pricing can change independently of LLMSwap.

Configure and verify

export OPENAI_API_KEY="your-key"
llmswap doctor --provider openai
llmswap providers

Make a real provider request only when appropriate:

llmswap providers --provider openai --verify

Choose a provider and model

Global CLI options appear before the command:

llmswap --provider anthropic ask "Review this proposal"

Set persistent defaults:

llmswap config set provider.default anthropic
llmswap config set provider.models.openai gpt-5.6
llmswap config show

Python:

from llmswap import LLMClient

client = LLMClient(provider="anthropic", model="claude-sonnet-5")
response = client.query("Review this proposal")

Explicit model IDs are passed through to the provider, but an ID can still fail when it is invalid, retired, regionally unavailable, or not enabled for the account.

Tool schemas

Tool schemas are implemented for Anthropic, OpenAI, Gemini, Groq, and xAI. Other provider adapters support text chat but not LLMSwap’s shared tool schema.

Ollama

Install Ollama separately, start its server, and pull the configured model:

ollama pull qwen3.5:9b
llmswap --provider ollama ask "Explain this locally"

llmswap doctor --provider ollama reports the exact pull command when the configured model is missing.

IBM watsonx

Install the optional dependency and export both credential fields:

python -m pip install "llmswap[watsonx]"
export WATSONX_API_KEY="your-key"
export WATSONX_PROJECT_ID="your-project-id"

Privacy and fallback

Remote prompts are sent to the chosen provider. If fallback is enabled, a failed request can move to another configured provider. Set fallback=False in Python when a request must stay on its selected provider.

Best Answer stays on the primary provider by default. Cross-provider synthesis requires explicit consent because candidate response text is sent to the judge provider.

For retirement migrations and official provider-source links, see the audited provider catalog.