heyllm

Talk to an LLM from your terminal. heyllm is a tiny, dependency-light CLI that streams answers from any OpenAI-compatible Chat Completions API straight to stdout — built for daily use and pipelines.

$ heyllm "explain monads simply"
A monad is a way to wrap a value along with a recipe for chaining...

It reads piped input as context, so it slots naturally into shell workflows:

$ git diff | heyllm "write a conventional commit message"
feat(parser): handle SSE frames split across network chunks

Install

npm install -g heyllm
# or
pnpm add -g heyllm

Requires Node.js ≥ 18 (it uses the built-in global fetch).

Setup

Set your API key once:

export OPENAI_API_KEY="sk-..."

If no key is found (and --api-key isn't passed), heyllm prints a clear error to stderr and exits with code 2.

Usage

heyllm [options] "your prompt"
command | heyllm [options] "instruction"

Options
  -m, --model <name>     Model to use (default: gpt-4o-mini)
      --system <text>    System prompt
      --base-url <url>   API base URL (default: https://api.openai.com/v1)
      --api-key <key>    API key (defaults to $OPENAI_API_KEY)
      --no-stream        Wait for the full response instead of streaming
      --json             Print the raw JSON response
  -h, --help             Show help
  -v, --version          Show version

Examples

# Stream an answer
heyllm "explain monads simply"

# Pipe context in; the prompt is the instruction, stdin is the context
git diff | heyllm "write a conventional commit message"

# Pick a model and add a system prompt
heyllm -m gpt-4o --system "You are terse" "summarize REST in 3 bullets"

# Get the raw API response for scripting
heyllm --json "ping" | jq '.usage'

# Disable streaming (single request, full response at once)
heyllm --no-stream "what is 2+2"

OpenAI-compatible endpoints

heyllm speaks the OpenAI Chat Completions wire format, so --base-url lets it talk to anything that implements it:

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

# Local Ollama
heyllm --base-url http://localhost:11434/v1 -m llama3 "hi"

Streaming is parsed from server-sent events (data: lines, ignoring [DONE]), and tokens are printed as they arrive.

Honestly

This is a deliberately tiny alternative to bigger tools like llm or vendor CLIs. No plugins, no config files, no chat history — just one command that pipes well. If you want a small, auditable binary that does the obvious thing, that's the whole pitch.

Development

pnpm install
pnpm run typecheck   # tsc --noEmit
pnpm run test        # vitest run (no network — fetch is mocked)
pnpm run build       # tsup -> dist/

The network call is isolated in src/request.ts; the message builder, SSE parser, and arg parser in src/index.ts are pure and unit-tested.

License

MIT © 2026 Abdulmunim Jemal