Providers

Arcana supports 33+ LLM providers through the models.dev catalog. You bring your own API keys — Arcana auto-detects which providers are available and lets you switch models mid-session.

How Providers Work

Set any provider's API key via environment variable. Arcana detects it automatically — no per-vendor configuration required.

# Set a key — Arcana detects the provider
export OPENAI_API_KEY=sk-...

# Verify detection
arcana doctor

# See all detected providers and models
arcana providers
arcana models

Provider Comparison

Key metrics for the most popular providers. See Model Recommendations for task-specific picks.

Tier 1 — Frontier Models

Provider	Best Models	Context	Strengths	Env Var
OpenAI	gpt-5.6-sol, gpt-5.4, gpt-4.1	1M	Best all-round, tool use, reasoning	`OPENAI_API_KEY`
Anthropic	claude-opus-5, claude-sonnet-5, claude-fable-5	1M	Long context, agentic coding, safety	`ANTHROPIC_API_KEY`
Google	gemini-3.6-flash, gemini-3.5-flash, gemini-3.1-pro	1M	Huge context, multimodal, fast, cheap	`GEMINI_API_KEY`

Tier 2 — Strong Alternatives

Provider	Best Models	Context	Strengths	Env Var
xAI	grok-4.5, grok-build-0.1	1M	Coding, science, math, real-time data	`XAI_API_KEY`
MoonshotAI	kimi-k3, coding-kimi-k3	1M	#1 coding arenas, 2.8T params, long-horizon	`KIMI_API_KEY`
DeepSeek	deepseek-v4-pro	varies	Very cheap, high-capability reasoning	`DEEPSEEK_API_KEY`
AWS Bedrock	claude, llama, mistral	varies	Enterprise, IAM auth, no egress	`AWS_ACCESS_KEY_ID`
Mistral	mistral-large, codestral	128K	Fast, European, open-weight options	`MISTRAL_API_KEY`
Alibaba	qwen3.8-max, qwen3-max	1M	2.4T params, full-stack dev, multilingual	`DASHSCOPE_API_KEY`

Tier 3 — Free & Self-Hosted

Provider	Best Models	Context	Strengths	Env Var
Ollama	llama3, codellama, mistral	varies	Local, free, no API key needed	Auto-detected
OpenRouter	all models via proxy	varies	Single key for 300+ models	`OPENROUTER_API_KEY`
Cloudflare	gpt-4.1, claude, gemini, llama	1M	Workers AI, edge inference, 200+ models	`CLOUDFLARE_API_TOKEN`
Together	llama, mixtral	varies	Fast inference, cheap	`TOGETHER_API_KEY`
Groq	llama, mixtral	varies	Ultra-fast inference	`GROQ_API_KEY`

All Supported Providers

Arcana auto-detects providers from environment variables. The full list includes:

OpenAIOPENAI_API_KEY

AnthropicANTHROPIC_API_KEY

Google GeminiGEMINI_API_KEY

xAIXAI_API_KEY

AWS BedrockAWS_ACCESS_KEY_ID

DeepSeekDEEPSEEK_API_KEY

MistralMISTRAL_API_KEY

OpenRouterOPENROUTER_API_KEY

OllamaLocal — no key needed

CloudflareCLOUDFLARE_API_TOKEN

Together AITOGETHER_API_KEY

GroqGROQ_API_KEY

ReplicateREPLICATE_API_TOKEN

CohereCOHERE_API_KEY

Hugging FaceHF_TOKEN

FireworksFIREWORKS_API_KEY

SambanovaSAMBANOVA_API_KEY

Novita AINOVITA_API_KEY

Run arcana doctor to see which keys are detected on your system.

Switching Providers

You can switch providers mid-session or configure defaults:

Environment Variable

# Set the default provider
export ARCANA_PROVIDER=anthropic
export ARCANA_MODEL=claude-sonnet-4

# Or use provider-specific keys
export ANTHROPIC_API_KEY=sk-ant-...

Config File

{
  "provider": "anthropic",
  "model": "claude-sonnet-4"
}

Mid-Session Switch

# In the TUI, type:
/providers
# Then select a different provider/model

Bring Your Own Key (BYOK)

Arcana works in two modes:

BYOK Mode (Free)

Set your own provider API keys (OPENAI_API_KEY, ANTHROPIC_API_KEY, etc.) and Arcana connects directly. You pay your providers directly — Arcana does not mark up costs or take a cut.

Arcana License Mode

For proxy, cloud sync, and Gateway features, you need an Arcana license. This is not a provider API key — it's a separate credential that unlocks Arcana's hosted infrastructure:

# Arcana license for proxy/cloud features
export ARCANA_LICENSE=your-license-key

# BYOK still works alongside your license
export OPENAI_API_KEY=sk-...

The proxy routes requests through Arcana's infrastructure using your license. BYOK mode (direct provider keys) is always free and requires no license.

Setup Guides

Step-by-step setup for each major provider:

OpenAI GPT-4o, o1, o3 — set OPENAI_API_KEY Anthropic Claude Sonnet, Opus — prompt caching Google Gemini 1M context, multimodal, free tier xAI Grok Real-time data, OpenAI-compatible AWS Bedrock IAM auth, enterprise, no egress DeepSeek Very cheap, strong coding Mistral Fast, European, Codestral OpenRouter 100+ models, single API key Ollama Fully local, free, no API key Cloudflare Workers AI Edge inference, serverless GPU, free tier Together AI Fast open-source inference, fine-tuning

Troubleshooting

Run arcana doctor to verify key detection
Keys must be set before starting Arcana
Some providers (Bedrock) require multiple env vars
See Troubleshooting for common issues

Last updated: Jul 24, 2026