CLI Companion Chat

`ed companion chat`

Talks with the companion. Retrieval grounds the reply in your own episodes, the reply streams to stdout as the model produces it, and validated citations print after it. Every exchange persists, so a conversation can be continued later from any machine.

Usage:

ed companion chat <message> [--conversation <id>] [--persona <id>] [--json] [--endpoint <url>]

Options:

Name	Type / values	Default	What it does
`--conversation`	conversation id	new conversation	Continues that conversation with its history in context.
`--persona`	persona id	`friend`	Answers through that lens; `ed companion personas` lists them.
`--json`	flag	off	Suppresses streaming and emits one JSON document at the end.
`--endpoint`	URL	resolution order	Uses this Companion API base URL.

Before starting the stream, --persona is checked against the live persona list when that list can be fetched. An unknown id exits 1 and names the known ids. If the persona list itself cannot be fetched, the chat request still gets a chance to report the backend error.

--json shape:

{
  "answer": "You shipped the auth refactor and kept avoiding the billing migration.",
  "chunksConsidered": 8,
  "citations": [
    {
      "episodeId": "6a7c1f0e-6f0f-4bb0-9d3a-2f6f6f0e1a2b",
      "occurredAt": "2026-08-05T09:00:00Z",
      "quote": "shipped the session-scoped tokens today",
      "support": "verbatim",
      "title": "auth-refactor.md"
    }
  ],
  "conversationId": "e3b6d2a4-27c8-4f7c-9b7e-3e2b1a0c9d8f",
  "latencyMs": 1874,
  "model": "openai-compatible at http://ollama:11434/v1, model qwen3:1.7b"
}

The conversation id prints on stderr after a plain-text chat; pass it back with --conversation to keep talking in the same thread.

Behaviour: requires a configured reasoning provider (ed companion reason); without one the backend answers 412 and the command exits 1. A transport failure, including no server at the endpoint, exits 4.