llmctl is a minimal, Unix-style command-line tool for interacting with Large Language Models (LLMs).

It is designed for ask-only workflows, where LLM output is treated like any other CLI program output: plain text via stdout, suitable for piping, redirection, and version control.

No agents.
No hidden state.
No browser UI.
No IDE lock-in.

llmctl follows classic Unix principles:

• Do one thing well
• Text in, text out
• Explicit context instead of implicit memory
• User stays in control

LLMs are treated as stateless text processors, not autonomous agents.

• Ask LLMs directly from the command line
• Streamed output to stdout
• Explicit context via files
• Works with OpenAI-compatible APIs
• Easy integration with editors, pipes, and scripts
• No automatic code execution
• No filesystem side effects unless you explicitly redirect output
• Estimated token counter

Build from source (Go):

git clone https://github.com/sebidev/llmctl.git
cd llmctl
go build -o llmctl

Place the binary somewhere in your PATH.

llmctl reads configuration from environment variables:

export OPENAI_API_KEY="your-api-key"
export OPENAI_BASE_URL="https://api.openai.com/v1"
export OPENAI_MODEL="gpt-4.1-mini"

This allows easy switching between cloud providers and local OpenAI-compatible servers (e.g. LM Studio, llama.cpp servers, etc.).

Simple prompt:

llmctl "explain systemd timers"

Redirect output to a file:

llmctl "hello world program in C++" > hello.cpp

Append to an existing document:

llmctl "add a summary section" >> notes.md

Provide explicit context using files:

llmctl --context notes.md "expand section 5"

Append the result back into the same file:

llmctl --context notes.md "expand section 5" >> notes.md

This makes the file itself the source of truth and conversation memory.

To avoid sending entire large files, limit context size:

llmctl --context notes.md --tail 2500 "continue this section"

Only the last N characters (or tokens, depending on implementation) are sent.

Because llmctl uses stdout, it works naturally with pipes:

cat error.log | llmctl "explain this error"

llmctl "list common nginx systemd units" | grep nginx

llmctl "generate markdown outline" >> draft.md

• Writing and extending Markdown documents
• Code generation and explanation
• Infrastructure notes and documentation
• Learning and research
• Shell-based workflows without browser or IDE integration

• Not an autonomous agent
• Not an IDE plugin
• Not a chat application
• Not a code executor

If you want automation, use scripts. If you want memory, use files. If you want control, use llmctl.

MIT License

llmctl is intentionally boring.

That is its strength.