llcat 0.7

/usr/bin/cat for LLMs

/usr/bin/cat for LLMs

llcat is an unambitious dumb program for LLM I/O.

That's why it's great!

You can handle this!

Very Quick Start

Got 0.3 seconds to spare?

List the models on OpenRouter:

uvx llcat -s https://openrouter.ai/api -m

Run that right now.

I'll wait.

---

llcat solves all your problems.

Yes. Even that one.

It can also:

• Pipe things from stdin and/or be prompted on the command line.
• Store conversation history optionally, as a boring JSON file.
• Do tool calling using the OpenAI spec. There's an example in this repository (and below).
• Use local or remote servers, authenticated or not.
• List models using -m without arguments. Specify a model with the argument.

Free Samples? Sure! It's Free Software.

• pipx install llcat
• uvx llcat

Dependencies? Just the requests library.

Note: It's llcat, not llmcat. Let's keep it pronounceable.

Examples

Start a chat with llama:

$ llcat -s https://openrouter.ai/api \
        -m meta-llama/llama-3.2-3b-instruct:free \
        -c /tmp/convo.txt \
        -k $(cat openrouter.key) \
        "What is the capital of France?"

Continue with Qwen:

$ llcat -s https://openrouter.ai/api \
        -m qwen/qwen3-4b:free \
        -c /tmp/convo.txt \
        -k $(cat openrouter.key) \
        "And what about Canada?"

And finish on the local network:

$ llcat -s http://192.168.1.21:8080 \
        -c /tmp/convo.txt \
        "And what about Japan?"

One conversation, hopping across models and servers.

Pure sorcery.

Summon Some More

Want to store state? Let's go!

$ source examples/fancy.sh
$ llc-server http://192.168.1.21:8080
$ llc "write a diss track where the knapsack problem hates on the towers of hanoi"

Go read the four lines of fancy.sh

Surprise! Environment variables and a wrapper function. That's all you need.

The Tool Call To Rule Them All

This example, a very strange way to play mp3s, uses the sophisticated 21 line tool_program.py included in this repository. Some people call it MCP, llcat calls it subprocess.run.

This demo also uses DA`/50's pretty little streaming markdown renderer, streamdown.

Kablam! Alright a16z where's my $50 million?

The enterprise applications are limitless...

Conversant Conversations?

Sure! Conversation.sh is six lines of bash.

It even has inspectable tool calls and realtime observability with Human-in-the-Loop! Wowza! I'll settle for merely $40 million.

Boring Documentation

usage: llcat [-h] [-c CONVERSATION] [-m [MODEL]] [-k KEY] [-s SERVER]
                [-tf TOOL_FILE] [-tp TOOL_PROGRAM]
                [prompt ...]

positional arguments:
  prompt                Your prompt

options:
  -h, --help            show this help message and exit
  -c, --conversation CONVERSATION
                        Conversation history file
  -m, --model [MODEL]   Model to use (or list models if no value)
  -k, --key KEY         API key for authorization
  -s, --server SERVER   Server URL (e.g., http://::1:8080)
  -tf, --tool_file TOOL_FILE
                        JSON file with tool definitions
  -tp, --tool_program TOOL_PROGRAM
                        Program to execute tool calls

Now it's your turn.

---

Brought to you by DA`/50: Make the future obvious.