Major Features / Model Providers

Model Providers

Heddle is not limited to OpenAI or Anthropic. It supports a broad family of hosted, local, self-hosted, and gateway-routed models so users can choose the model stack that fits their workflow.

Supported provider families

Provider family	Model prefix	Access mode	Model discovery	Smoke test
OpenAI	`gpt-`, `o`	OpenAI account sign-in or `OPENAI_API_KEY`	Built-in curated shortlist	Normal `heddle --model ... ask` run
Anthropic Claude	`claude-*`	`ANTHROPIC_API_KEY`	Built-in curated shortlist	Normal `heddle --model ... ask` run
Ollama	`ollama/` or `ollama:`	Local endpoint, no hosted key required	Native Ollama `/api/tags`	`yarn smoke:ollama`
LM Studio	`lmstudio/`	Local OpenAI-compatible endpoint	`/models` when the server is running	`yarn smoke:lmstudio`
LiteLLM	`litellm/`	OpenAI-compatible gateway, optional gateway key	`/models` when the gateway is reachable	`yarn smoke:litellm`
vLLM	`vllm/`	Self-hosted OpenAI-compatible endpoint, optional server key	`/models` when the server is running	`yarn smoke:vllm`
Hugging Face router	`huggingface/` or `hf/`	`HF_TOKEN` or `HUGGINGFACE_API_KEY`	`/models` when the API key is configured	`yarn smoke:huggingface`
OpenRouter	`openrouter/`	`OPENROUTER_API_KEY`	`/models` when the API key is configured	`yarn smoke:openrouter`
Together AI	`together/`	`TOGETHER_API_KEY`	`/models` when the API key is configured	`yarn smoke:together`
Groq	`groq/`	`GROQ_API_KEY`	`/models` when the API key is configured	`yarn smoke:groq`

Model selection examples

heddle --model gpt-5.4-mini ask "Summarize this repository"
heddle --model claude-sonnet-4-6 ask "Summarize this repository"
heddle --model ollama/llama3.2:latest ask "Summarize this repository"
heddle --model lmstudio/local-model ask "Summarize this repository"
heddle --model openrouter/meta-llama/llama-3.3-70b-instruct ask "Summarize this repository"

/model set llama
/model ollama/llama3.2:latest
/model openrouter/meta-llama/llama-3.3-70b-instruct

Quality caveat

Provider compatibility means Heddle can route requests and tool definitions to that provider. It does not mean every model behind that provider is equally good at coding-agent work.

Smaller, older, heavily quantized, local, or aggressively routed models may miss tool calls.
Some models call tools correctly but ignore the tool result.
Some models answer confidently with incorrect repository facts.
Slow local runs can hit request timeouts.

Related setup

For environment variables, auth commands, default models, and picker behavior, read Providers & Models.

Providers & Models