WARING: gptel-watch.el Most of the code is written by AI.

gptel-watch Inspired by aider --watch-files

gptel-watch is an Emacs minor mode that automatically invokes gptel-request when the user types certain lines that indicate intent — for example, lines ending with "AI!" or "#ai". It uses GPT-style large language models to assist writing or coding directly within your buffer, based on local context.

This package builds on gptel, providing a lightweight, pattern-driven workflow for triggering GPT responses while typing.

• Monitors user input and triggers GPT when a line ends with defined patterns.
• Custom extraction of context, see title.: How to Context extraction
• Clears the trigger line and replaces it with GPT-generated output.
• Works seamlessly with newline, org-return, or any custom command you define.
• Supports global and per-buffer activation.

(use-package gptel-watch
  :after gptel
  :load-path "/path/gptel-watch/"
  :config
  (gptel-watch-global-mode 1))  ;; Optional: enable globally

Once enabled, write a line like this:

// Print Hello World. AI!

Then press RET. The line will be replaced with the GPT-generated result, such as:

printf("Hello World");

1. Current Defun: Current definition under cursor
2. Down/Up Line: Up/Down a line
3. Line Range: Precise line number range of the current buffer
4. Only Current Line: Only the current line, do not extract context.

All customization can also be set via setq in your init.el

1. Write your idea or prompt inline with a marker like // AI! or #ai.

2. Press RET.

3. GPT fills in the next logical content based on surrounding context.

4. Continue writing!

This package uses post-command-hook to monitor input commands.

• Only non-minibuffer buffers are affected.

• Trigger pattern matching is based on line suffixes — regexes are appended with $ internally.

• The system prompt is critical for shaping the LLM output — tune it to suit your use case (e.g., short replies, no Markdown, coding only, etc.).