llms

A tiny Rust CLI for streaming responses from OpenAI, Gemini, and Claude.

What it does

• Streams model output to stdout as tokens/chunks arrive.
• Reads prompt from --prompt or stdin.
• Supports provider API keys from env vars or a home config file.
• Includes config commands:
  • llms config set <provider> <key>
  • llms config get <provider>
  • llms config list

Requirements

• Rust toolchain (stable)
• Network access to the model provider API
• An API key for the provider you use

Build

cargo build --release

Binary path:

./target/release/llms

Or run directly with cargo:

cargo run -- --help

Configure API keys

You can use either env vars or the built-in config file.

Option A: env vars

export OPENAI_API_KEY="sk-..."
export GEMINI_API_KEY="..."
export ANTHROPIC_API_KEY="..."

Option B: CLI config commands

llms config set openai sk-xxxx
llms config set gemini your-gemini-key
llms config set claude your-anthropic-key

Read a key:

llms config get openai

List configured providers:

llms config list

Config file location:

~/.llms/config.json

Example file:

{
  "openai": "sk-xxxx",
  "gemini": "your-gemini-key",
  "claude": "your-anthropic-key"
}

Key precedence:

1. Provider env var (if set and non-empty)
2. ~/.llms/config.json

Usage

Top-level help:

llms --help

Basic prompt

llms --provider openai --model gpt-5 --prompt "Explain SSE in one sentence"

Read prompt from stdin

echo "Write a haiku about rust" | llms --provider gemini --model gemini-3-flash-preview

Provider defaults

If --model is omitted, defaults are:

• openai: gpt-5
• gemini: gemini-2.0-flash
• claude: claude-opus-4-6

Piping and redirection

The CLI writes model text to stdout and errors to stderr, so normal shell piping works.

Save output to file:

llms --provider gemini --prompt "hello" > out.txt

Pipe to another process:

llms --provider gemini --prompt "hello" | tee out.txt

Split output and errors:

llms --provider gemini --prompt "hello" > out.txt 2> err.log

If running through cargo and you do not want cargo status lines:

cargo run --quiet -- --provider gemini --prompt "hello"

Troubleshooting

• Missing API key for <provider>
  • Set the matching env var, or run llms config set <provider> <key>.
• No response when using stdin
  • Ensure stdin is not empty and not just whitespace.
• Wrong provider name
  • Valid providers are openai, gemini, claude.

Notes

• API keys are stored in plain text in ~/.llms/config.json.
• Treat that file as sensitive.

License

MIT