Pipe text to LLMs from the terminal — while you still can

What it does

Mods reads standard input, prepends your prompt arguments, and sends the text to an LLM, printing the response so you can transform command output into Markdown, JSON, or plain text. It acts as a pipeline filter, working with local models through LocalAI or hosted APIs including OpenAI, Groq, Cohere, Azure, and Gemini.

The interesting bit

Conversations are saved locally with SHA-1 identifiers and titles—styled after git commits—and can be resumed or reviewed later. The tool also supports custom YAML-defined roles, such as a “shell” role that emits only one-liner commands without explanation, and recently added MCP server integration.

Key highlights

• Built for pipelines: consumes stdin and writes formatted responses to stdout
• Supports multiple backends: LocalAI, OpenAI, Cohere, Groq, Azure OpenAI, and Gemini
• Persistent conversations with SHA-1 IDs and continuation flags
• Custom system prompts via YAML roles
• MCP server listing and tool discovery

Caveats

• The project is officially sunsetting on March 9, 2026; Charm is shifting focus to Crush and its non-interactive crush run mode
• Much of the current functionality is expected to migrate to Crush, so new feature investment here has ended

Verdict

Grab it if you want a lightweight, pipe-friendly LLM client today and don’t mind migrating later—or if you’re interested in maintaining a fork. Otherwise, keep an eye on Crush’s non-interactive mode.