🌎 Create client-side version and host on `chainforge.ai`

[ianarawjo]

NOTE: We are currently working on this major change in the branch js-backend.

Many users have requested being able to use ChainForge directly from the browser, without pip install. To host ChainForge, however, requires significant changes.

The cleanest, least-security-risk way to host ChainForge is to convert the entire Python backend into client-side Javascript (flask_app.py, query.py, etc). This requires the following functional differences, among many other changes:

• the cache of LLM responses is limited to the currently opened flow
• add a Javascript Evaluator Node which runs in the browser client-side, and remove the Python Evaluator Node for setups that use the Javascript 'backend'

The client-side version has a few drawbacks as a consequence:

• limitations of cacheing in the browser (~5MB on localStorage, meaning only autosaves apply)
• cannot easily use imports in client-side Javascript, preventing access to third-party libraries that can help with evaluation
• removing access to Python Evaluation Node, which again removes a nice use case of importing Python modules or evaluating Python code

Thus, the current goal is to retain the Python Flask backend when users are running ChainForge locally. When users are running from Flask (locally), they will be able to still:

• add and access a Python Evaluator Node, as before
• run Python evaluator functions on their machine's backend, as before,
• cache responses on their local machine in the cache/ folder, as before,

all while most of the logic and API calls to model providers (e.g., generating prompt permutations from templates, calling OpenAI) is moved to the front-end.

[ianarawjo]

Part of the reason we want to do this is to add a 'share' button as well, so that, ideally, users on chainforge.ai can click a Share button which creates a unique link to their flow (at first, due to data limitations and concerns, likely without the cache of LLM responses). That would require an additional issue:

• add a server-side database for storing user flows (e.g. with PHP and MySQL)
• rate-limit user's ability to 'share' their flow (for security concerns)
• add a server-side routing mechanism that given a URL `chainforge.ai/play/', fetches the saved JSON for that flow (if it exists) and loads it into the front-end app

[ianarawjo]

This change has largely been performed, and is currently undergoing testing. Note that:

• cache responses on their local machine in the cache/ folder, as before,

is no longer a goal --the cache folder will no longer be used. Instead, all results will stay local to a given flow. If user wish to 'save/load' a flow, users will need to click export/import buttons, and the cforge file will contain the relevant cache info.

[ianarawjo]

The browser based version of ChainForge has been pushed to main and PyPI. A version will be hosted on chainforge.ai on Wednesday July 5th.

[ianarawjo]

This issue is currently resolved. See the website for details.