A provider is a model API: Anthropic, OpenAI, Gemini, a local Ollama server, any OpenAI-compatible service. Providers live under providers: in config.yaml, and an agent names its model as <provider>/<model id>.

yaml

providers:
  anthropic: {}               # a provider named after a preset uses it
  local:
    preset: ollama
agents:
  habakkuk:
    model: anthropic/claude-opus-5-5
    fallbacks: [local/llama3.3]

habakkuk models lists each provider with its credential status and, where the API can list them, its models. Nothing in Habakkuk assumes one vendor.

In this section#

Providers provider types, presets, custom and local APIs, key rotation and fallbacks.Routing a cheap model for quick messages, a strong one for real work.Decision model a fast yes/no model for Habakkuk's own choices, such as routing and routine approvals.Subscription logins use a ChatGPT or GitHub Copilot plan instead of an API key.Usage and limits tokens and cost per model and agent, and hard spending caps.Speech transcription for voice notes, and text-to-speech for spoken replies.

Provider directory#

Each preset fills in the API type, base URL and key variable. Set preset:, or name the provider after the preset. Names in brackets work too.

Preset	Base URL	Key variable
`anthropic`	Anthropic API	`ANTHROPIC_API_KEY`
`openai`	`https://api.openai.com/v1`	`OPENAI_API_KEY`
`gemini` (`google`)	Gemini API (type `gemini`)	`GEMINI_API_KEY`
`bedrock`	AWS (type `bedrock`)	AWS credential chain
`openrouter`	`https://openrouter.ai/api/v1`	`OPENROUTER_API_KEY`
`groq`	`https://api.groq.com/openai/v1`	`GROQ_API_KEY`
`together`	`https://api.together.ai/v1`	`TOGETHER_API_KEY`
`fireworks`	`https://api.fireworks.ai/inference/v1`	`FIREWORKS_API_KEY`
`deepseek`	`https://api.deepseek.com`	`DEEPSEEK_API_KEY`
`xai` (`grok`)	`https://api.x.ai/v1`	`XAI_API_KEY`
`mistral`	`https://api.mistral.ai/v1`	`MISTRAL_API_KEY`
`cerebras`	`https://api.cerebras.ai/v1`	`CEREBRAS_API_KEY`
`deepinfra`	`https://api.deepinfra.com/v1/openai`	`DEEPINFRA_API_KEY`
`nvidia` (`nim`)	`https://integrate.api.nvidia.com/v1`	`NVIDIA_API_KEY`
`huggingface` (`hf`)	`https://router.huggingface.co/v1`	`HF_TOKEN`
`perplexity`	`https://api.perplexity.ai`	`PERPLEXITY_API_KEY`
`moonshot` (`kimi`)	`https://api.moonshot.ai/v1`	`MOONSHOT_API_KEY`
`zai` (`glm`)	`https://api.z.ai/api/paas/v4`	`ZAI_API_KEY`
`minimax`	`https://api.minimax.io/v1`	`MINIMAX_API_KEY`
`qwen` (`dashscope`)	`https://dashscope-intl.aliyuncs.com/compatible-mode/v1` (keys are region-bound; set `base_url` for your region)	`DASHSCOPE_API_KEY`
`ollama`	`http://localhost:11434/v1`	optional `OLLAMA_API_KEY`
`lmstudio`	`http://localhost:1234/v1`	optional `LM_API_TOKEN`
`vllm`	`http://localhost:8000/v1`	optional `VLLM_API_KEY`
`llamacpp` (`llama.cpp`)	`http://localhost:9931/v1` (older llama-server builds use 8080)	optional `LLAMA_API_KEY`
`litellm`	`http://localhost:4000/v1`	optional `LITELLM_API_KEY`
`vercel` (`vercel-ai-gateway`)	`https://ai-gateway.vercel.sh/v1`	`AI_GATEWAY_API_KEY`
`cloudflare` (`cloudflare-ai-gateway`)	set `base_url: https://gateway.ai.cloudflare.com/v1/ACCOUNT_ID/GATEWAY_ID/compat`	`CLOUDFLARE_AI_GATEWAY_KEY`

Any field you set wins over the preset. A service that is not listed works too, if it speaks one of the provider types.