Configuration

heyllm has no config files. Everything it needs comes from a couple of flags and one environment variable — which is the whole point of a tiny, auditable tool. There’s nothing to discover on disk, nothing to migrate, and nothing hidden.

Authentication

heyllm needs an API key. It looks in two places, in order:

1. The --api-key <key> flag.
2. The OPENAI_API_KEY environment variable.

The usual setup is to export the variable once (e.g. in your shell profile):

export OPENAI_API_KEY="sk-..."
heyllm "hello"

Pass --api-key to override it for a single call — handy when talking to a provider other than OpenAI without changing your environment:

heyllm --api-key "$OPENROUTER_API_KEY" "hello"

If no key is found by either route, heyllm prints a clear error to stderr and exits with code 2. The key is sent as a standard Authorization: Bearer <key> header.

Choosing a model

The default model is gpt-4o-mini. Use -m / --model to pick another:

heyllm -m gpt-4o "summarize REST in 3 bullets"

The model name is passed straight through to the API, so the valid values are whatever your chosen provider accepts. With OpenRouter, for example, names are namespaced like openai/gpt-4o-mini or anthropic/claude-3.5-sonnet; with a local Ollama they’re whatever you’ve pulled (llama3, mistral, …).

OpenAI-compatible providers

heyllm speaks the OpenAI Chat Completions wire format and posts to <base-url>/chat/completions. The default base URL is https://api.openai.com/v1, but --base-url lets it talk to anything that implements the same API. Any trailing slashes on the base URL are normalized before /chat/completions is appended.

OpenRouter

A single gateway in front of many providers — pass its base URL, your OpenRouter key, and a namespaced model name:

heyllm --base-url https://openrouter.ai/api/v1 \
    --api-key "$OPENROUTER_API_KEY" \
    -m openai/gpt-4o-mini "hi"

Local Ollama

A local Ollama server exposes an OpenAI-compatible endpoint on /v1. It doesn’t check the key, but heyllm still requires some value to be present — so set a placeholder if you don’t already have OPENAI_API_KEY exported:

export OPENAI_API_KEY="ollama"   # any non-empty value
heyllm --base-url http://localhost:11434/v1 -m llama3 "hi"

Anything else

The same pattern works for any OpenAI-compatible endpoint — Azure OpenAI, Together, Groq, a self-hosted vLLM, and so on. Point --base-url at the provider’s /v1 root, supply the matching --api-key, and pick a model name it recognizes.

Streaming behavior

By default heyllm streams tokens as they arrive, parsed from the provider’s server-sent events (data: lines, ignoring the final [DONE] sentinel). If a provider doesn’t support streaming — or you simply prefer to wait for the whole answer — pass --no-stream:

heyllm --no-stream --base-url https://some-provider/v1 "hello"

Raw output for scripting

When you need the full API payload — for example to read token usage — use --json to print the raw response instead of just the text. --json implies a non-streaming request:

heyllm --json "ping" | jq '.usage'

Configuration at a glance

What	How	Default
API key	--api-key or $OPENAI_API_KEY	(none — required)
Model	-m / --model	gpt-4o-mini
Endpoint	--base-url	https://api.openai.com/v1
System prompt	--system	(none)
Streaming	on by default; --no-stream to disable	on