llm-prompt-json

A plugin for LLM that adds a --json flag to the llm prompt command. This flag allows you to retrieve the response, conversation ID, and metadata as a JSON array, making it easy to integrate with tools like jq.

Built for llm version 0.28.

Installation

Install this plugin in the same environment as LLM:

llm install llm-prompt-json

Usage

Use the --json flag when running a prompt:

llm "say hello" --json

Example output:

[
  {
    "id": "01...",
    "model": "gpt-4o",
    "prompt": "say hello",
    "response": "Hello!",
    "conversation_id": "01..."
  }
]

Extracting the Conversation ID

Because the output is standard JSON, you can easily pipe it into jq to extract just the values you need. For example, to get the conversation_id:

llm "say hello" --json | jq -r '.[].conversation_id'

Compatibility

This plugin intercepts llm prompt dynamically, preventing the usual streamed output from polluting the JSON payload. It works correctly even if streaming is enabled or disabled, but it cannot be used alongside the --no-log flag since it relies on the internal LLM logs database to construct the complete metadata record.