Provider Configuration¶

Providers are the AI model backends that Missy uses for inference. Each provider is configured as a named entry under the providers section.

Basic Configuration¶

providers:
  anthropic:
    name: anthropic
    model: "claude-sonnet-4-6"
    timeout: 30
    enabled: true

The only required field is model. All other fields have sensible defaults.

Provider Fields¶

API Key Resolution¶

Missy resolves API keys using a priority chain:

flowchart TD
    A[api_key field] --> B{Set in config?}
    B -->|Yes| C{vault:// or $ENV?}
    C -->|Yes| D[Resolve reference]
    C -->|No| E[Use literal value]
    B -->|No| F{api_keys list has entries?}
    F -->|Yes| G[Use first entry from api_keys]
    F -->|No| H[Check env var: PROVIDER_API_KEY]
    H --> I{Env var set?}
    I -->|Yes| E
    I -->|No| J[No API key — provider may fail]

1. api_key in the config file (literal, vault://, or $ENV reference).
2. First entry from api_keys list (if api_key is not set).
3. Environment variable {PROVIDER_NAME}_API_KEY (uppercased).

Vault References¶

Store API keys in Missy's encrypted vault instead of plain text in the config file:

providers:
  anthropic:
    api_key: "vault://ANTHROPIC_API_KEY"

Set the vault value:

missy vault set ANTHROPIC_API_KEY "sk-ant-..."

Both api_key and individual entries in api_keys support vault:// references. Environment variable references with $ prefix are also resolved:

providers:
  openai:
    api_key: "$OPENAI_API_KEY"    # resolves from environment

API Key Rotation¶

The api_keys field accepts a list of keys. The provider implementation rotates through them when rate limits are hit:

providers:
  anthropic:
    model: "claude-sonnet-4-6"
    api_keys:
      - "vault://ANTHROPIC_KEY_1"
      - "vault://ANTHROPIC_KEY_2"
      - "vault://ANTHROPIC_KEY_3"

Model Tiers¶

Each provider can define three model tiers for different workload types:

providers:
  anthropic:
    model: "claude-sonnet-4-6"         # default tier
    fast_model: "claude-haiku-4-5"       # fast/cheap tier
    premium_model: "claude-opus-4-6"     # premium/complex tier

The agent runtime and model router select the appropriate tier based on task complexity.

Base URL Override¶

For self-hosted models or API-compatible proxies, override the endpoint:

providers:
  ollama:
    name: ollama
    model: "llama3.2"
    base_url: "http://localhost:11434"
    timeout: 120

providers:
  custom:
    name: openai
    model: "gpt-4o"
    base_url: "https://my-proxy.internal.corp/v1"
    api_key: "vault://PROXY_API_KEY"

providers:
  azure:
    name: openai
    model: "gpt-4o"
    base_url: "https://my-deployment.openai.azure.com"
    api_key: "vault://AZURE_OPENAI_KEY"
    timeout: 60

Timeout Configuration¶

The timeout field sets the HTTP request timeout in seconds. Adjust based on the provider and model:

Disabling a Provider¶

Set enabled: false to disable a provider without removing its configuration:

providers:
  openai:
    model: "gpt-4o"
    api_key: "vault://OPENAI_API_KEY"
    enabled: false                       # configured but not active

Disabled providers are skipped by the ProviderRegistry during resolution and fallback.

Provider Fallback¶

The ProviderRegistry resolves providers in this order:

1. The provider specified by --provider flag or AgentConfig.provider.
2. If that provider is unavailable (disabled, no API key, etc.), fall back to the first available provider in the registry.

Network Requirements¶

Each provider needs network access to its API endpoint. Use network presets for easy configuration:

network:
  default_deny: true
  presets:
    - anthropic     # allows api.anthropic.com
    - openai        # allows api.openai.com
    - ollama        # allows localhost:11434

Multiple Provider Example¶

providers:
  anthropic:
    name: anthropic
    model: "claude-sonnet-4-6"
    fast_model: "claude-haiku-4-5"
    premium_model: "claude-opus-4-6"
    api_key: "vault://ANTHROPIC_API_KEY"
    timeout: 30

  openai:
    name: openai
    model: "gpt-4o"
    api_key: "vault://OPENAI_API_KEY"
    timeout: 30

  ollama:
    name: ollama
    model: "llama3.2"
    base_url: "http://localhost:11434"
    timeout: 120

network:
  default_deny: true
  presets:
    - anthropic
    - openai
    - ollama