feat: allow users to chat with apis

[mamadoudicko]

You can now create a brain which can fetch data from external APIs with or without authentification

• POST query example with authentification

Screen.Recording.2023-11-09.at.14.35.34.mp4

• Get query example with authentification and search params

Screen.Recording.2023-11-09.at.14.38.53.mp4

• Get query without authentification and search params

Screen.Recording.2023-11-09.at.14.37.27.mp4

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
docs	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Nov 9, 2023 1:53pm
quivr-strapi	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Nov 9, 2023 1:53pm
quivrapp	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Nov 9, 2023 1:53pm

[github-actions]

Risk Level 3 - /home/runner/work/quivr/quivr/backend/routes/chat_routes.py

1. Potential Security Issue: The current_user.openai_api_key is being set from the request headers. This could potentially be a security risk if the API key is not properly validated or sanitized. Consider validating and sanitizing the API key before using it.

2. Code Duplication: The code for setting the chat_question attributes (model, temperature, max_tokens) is duplicated in create_question_handler and create_stream_question_handler. Consider refactoring this into a separate function to adhere to the DRY (Don't Repeat Yourself) principle.

3. Error Handling: There is a bare except HTTPException as e block which simply re-raises the exception. This is unnecessary and can be removed. If additional error handling is required, it should be implemented in this block.

4. Code Readability: The is_model_ok check is done twice in the create_question_handler function. This could be confusing and should be simplified.

Example code snippet for the second point:

def set_chat_question_attributes(chat_question, brain):
    chat_question.model = chat_question.model or brain.model or \"gpt-3.5-turbo\"
    chat_question.temperature = chat_question.temperature or brain.temperature or 0.1
    chat_question.max_tokens = chat_question.max_tokens or brain.max_tokens or 512
    return chat_question

Then replace the duplicated code with a call to this function:

chat_question = set_chat_question_attributes(chat_question, brain)

---

Risk Level 3 - /home/runner/work/quivr/quivr/backend/llm/api_brain_qa.py

1. The make_completion method is quite long and complex, which makes it hard to understand. Consider breaking it down into smaller, more manageable functions.
2. The try/except block in the make_completion method is too broad. It's generally a bad practice to catch all exceptions. Instead, catch specific exceptions that you expect might be raised in the try block.
3. The generate_stream method is also quite long and complex. Consider breaking it down into smaller, more manageable functions.

Example:

try:
    arguments = json.loads(function_call[\"arguments\"])
except json.JSONDecodeError:
    arguments = {}

This way, you're only catching the exception that gets raised if json.loads fails to decode function_call[\"arguments\"].

---

Risk Level 3 - /home/runner/work/quivr/quivr/backend/repository/brain/create_brain.py

1. The create_brain function is not handling potential errors that could occur when creating the brain. Consider adding error handling to this function.

Example:

try:
    created_brain = supabase_db.create_brain(brain)
except Exception as e:
    logger.error(f'Error creating brain: {e}')
    raise

2. The create_brain function is not validating the brain_secrets_values dictionary. Consider adding validation to ensure that the dictionary contains valid data.

Example:

for secret_name, secret_value in secrets_values.items():
    if not isinstance(secret_name, str) or not isinstance(secret_value, str):
        raise ValueError('Invalid secret name or value')
    create_secret(
        user_id=user_id,
        brain_id=created_brain.brain_id,
        secret_name=secret_name,
        secret_value=secret_value,
    )

---

🔒🔁🔧

---

Powered by Code Review GPT

[StanGirard]

T'es une machine <3 @gozineb vous êtes chauds !

[gozineb]

Damn, awesome PR ! Really cool demo too 🔥 🔥 🔥 🔥 🔥 🔥 🔥

[mamadoudicko]

Thanks guys !! 🥳