Inline chat

Editors use the chat/inlinePrompt protocol method to ask ECA from any buffer and render the answer inline (e.g. an overlay near the cursor). The session is a regular chat behind the scenes: tool calls, approvals and follow-ups work as usual, and it can fork an existing chat to reuse its history as context.

Configure the model, agent, and variant used for inline chats via the chatInline config. By default, ECA follows the same selection as chat (or the source chat when forking) unless overwritten:

Example

~/.config/eca/config.json

{
  "chatInline": {
    "model": "github-copilot/gpt-4.1",
    "agent": "code",
    "variant": "low"
  }
}

model and variant apply when the inline chat is created (explicit client choice > chatInline config > source chat > default) and stick to the chat afterwards. agent applies to every inline prompt where the client doesn't specify one.