Providers

Anode names models as provider/model. Set one key and Anode registers that provider. Add config when you need custom endpoints, model lists, timeouts, or provider-specific behavior.

anode -x --model anthropic/claude-sonnet-4-6 "hello"
anode -x --model openai/gpt-5.5 "hello"

Fast Setup

Set one key:

export ANTHROPIC_API_KEY="..."
anode

Or use OpenAI login:

anode login

Open the model picker in the TUI:

/model

Auto-Detected Providers

Provider ID	Environment variable	Built-in model list starts with
anthropic	ANTHROPIC_API_KEY	claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5
openai	OPENAI_API_KEY	gpt-5.5, gpt-5.5-pro, gpt-5.4
zai	ZAI_API_KEY	glm-5.1, glm-5, glm-5-turbo
zai-coding	ZAI_CODING_API_KEY	GLM-5.1, GLM-5, GLM-5-Turbo
gemini	GEMINI_API_KEY	gemini-2.5-flash, gemini-2.5-pro
groq	GROQ_API_KEY	llama-3.3-70b-versatile, llama-3.1-8b-instant
deepseek	DEEPSEEK_API_KEY	deepseek-v4-pro, deepseek-v4-flash
mistral	MISTRAL_API_KEY	mistral-large-latest, codestral-latest
fireworks	FIREWORKS_API_KEY	Fireworks model paths
xai	XAI_API_KEY	grok-4.3, grok-4.20-0309-reasoning
openrouter	OPENROUTER_API_KEY	pass-through OpenRouter model names
together	TOGETHER_API_KEY	Together model names
cerebras	CEREBRAS_API_KEY	llama-4-scout-17b-16e-instruct, qwen3-32b
deepinfra	DEEPINFRA_API_KEY	DeepInfra model names
huggingface	HUGGINGFACE_API_KEY	Hugging Face Inference model names
ollama	OLLAMA_HOST	llama3.3, qwen3, codestral

Default model selection prefers anthropic/claude-sonnet-4-6, then openai/gpt-5.5, then zai-coding/GLM-5.1, then zai/glm-5.1.

Configured Providers

Add providers in ~/.config/anode/config.json or the selected $ANODE_CONFIG:

{
  "providers": {
    "my-llm": {
      "type": "openai-compatible",
      "apiKey": "{env:MY_LLM_KEY}",
      "baseURL": "https://llm.example.com/v1",
      "models": ["fast", "large"]
    }
  },
  "model": "my-llm/fast"
}

Provider fields:

Field	Use
type	Provider type. If omitted, many built-in provider IDs work as their own type.
apiFormat	Custom endpoint wire format: openai, anthropic, gemini, or generic. Used with type: "custom".
apiKey	Literal key, {env:VAR}, bare env var name, or !shell command.
baseURL	Endpoint override. Required for most custom providers.
models	Model names Anode should offer for this provider.
modelDefaults	Per-model BYOK defaults (display name, reasoning capability, thinking budgets, token limits, vision support). See Configuration → Per-Model Defaults. The legacy modelCapabilities name still works.
timeoutSec	Request timeout. Default 120.
maxRetries	Retry count. Default 2; negative disables retries.

Retries apply to retryable HTTP responses, including rate limits and server errors. Anthropic, OpenAI-compatible, OpenAI Responses, and Gemini-native providers honor bounded Retry-After seconds or HTTP-date headers before falling back to exponential retry delays. When a run retries and then succeeds, the harness ledger records provider retry telemetry so anode harness inspect and anode harness ledger <run-id> can show the hidden backoff instead of making the run look clean but slow.

OpenAI-Compatible Endpoints

Use openai-compatible for vLLM, LiteLLM, LM Studio, SGLang, and similar servers:

{
  "providers": {
    "local": {
      "type": "openai-compatible",
      "baseURL": "http://localhost:8000/v1",
      "models": ["default"]
    }
  },
  "model": "local/default"
}

For local endpoints, omit apiKey if the server does not require one.

custom is similar, but keeps the configured provider ID and display name:

{
  "providers": {
    "lab": {
      "type": "custom",
      "apiKey": "{env:LAB_KEY}",
      "baseURL": "https://lab.example.com/v1",
      "models": ["lab-large"]
    }
  }
}

Custom providers can also select a non-OpenAI wire format while keeping the same endpoint picker in /connect:

{
  "providers": {
    "claude-proxy": {
      "type": "custom",
      "apiFormat": "anthropic",
      "apiKey": "{env:CLAUDE_PROXY_KEY}",
      "baseURL": "https://claude-proxy.example.com",
      "models": ["claude-sonnet-4-6"]
    },
    "gemini-proxy": {
      "type": "custom",
      "apiFormat": "gemini",
      "apiKey": "{env:GEMINI_PROXY_KEY}",
      "baseURL": "https://generativelanguage.googleapis.com/v1beta",
      "models": ["gemini-2.5-pro"]
    }
  }
}

generic is a conservative Chat Completions variant for endpoints that reject OpenAI-only extras such as developer-role messages or stream options.

Ollama

export OLLAMA_HOST="http://localhost:11434"
anode -x --model ollama/llama3.3 "hello"

If OLLAMA_HOST is not set and you do not configure an Ollama provider, Anode does not register Ollama.

Azure OpenAI And Bedrock

These are config-only providers:

{
  "providers": {
    "azure": {
      "type": "azure-openai",
      "apiKey": "{env:AZURE_OPENAI_KEY}",
      "baseURL": "https://example.openai.azure.com/openai/deployments/gpt-4o",
      "models": ["gpt-4o"]
    },
    "bedrock": {
      "type": "bedrock",
      "apiKey": "{env:BEDROCK_KEY}",
      "baseURL": "https://bedrock-runtime.us-east-1.amazonaws.com",
      "models": ["anthropic.claude-sonnet-4"]
    }
  }
}

Reasoning Effort

Some models expose reasoning effort. Anode maps the profile effort level to the provider-specific parameter and clamps it to the model capability range.

Use the TUI command:

/effort

Profiles set their default effort with defaultEffort.

If you connect to a custom or proxy provider that exposes models the built-in registry doesn’t recognize (private aliases, preview names, in-house BYOK wrappers), declare per-model defaults under providers[*].modelDefaults so the model picker shows a friendly label, /effort and reasoning controls work, and the agent assembles prompts within the right context/output budgets. See Configuration → Per-Model Defaults.

Troubleshooting

• provider not found means the provider did not register. Check the env var or config entry.
• model not found means the selected model is not in that provider’s model list.
• If a config provider has apiKey already set, environment auto-detection does not overwrite it.

Keep going:

• Configuration
• Profiles
• Troubleshooting