LLM_OPENAI_SHOW_RESPONSES=1 debug trick, closes #286

docs/contributing.md

@@ -18,6 +18,16 @@ To run the tests:
 
     pytest
 
+## Debugging tricks
+
+The default OpenAI plugin has a debugging mechanism for showing the exact responses that came back from the OpenAI API.
+
+Set the `LLM_OPENAI_SHOW_RESPONSES` environment variable like this:
+```bash
+LLM_OPENAI_SHOW_RESPONSES=1 llm -m chatgpt 'three word slogan for an an otter-run bakery'
+```
+This will output the response (including streaming responses) to standard error, as shown in [issues 286](https://github.com/simonw/llm/issues/286).
+
 ## Documentation
 
 Documentation for this project uses [MyST](https://myst-parser.readthedocs.io/) - it is written in Markdown and rendered using Sphinx.

llm/default_plugins/openai_models.py

@@ -4,6 +4,7 @@ from llm.utils import dicts_to_table_string
 import click
 import datetime
 import openai
+import os
 
 try:
     from pydantic import field_validator, Field  # type: ignore
@@ -15,6 +16,15 @@ from typing import List, Iterable, Iterator, Optional, Union
 import json
 import yaml
 
+if os.environ.get("LLM_OPENAI_SHOW_RESPONSES"):
+
+    def log_response(response, *args, **kwargs):
+        click.echo(response.text, err=True)
+        return response
+
+    openai.requestssession = requests.Session()
+    openai.requestssession.hooks["response"].append(log_response)
+
 
 @hookimpl
 def register_models(register):