Users should be able to export prompt calls to code

[ianarawjo]

Users should be able to export the prompt API calls data into python code for a prompt node, for use in other programs. (Either add … button -> “Export API calls to Python code” or add this to right-click context menu). Downloads self-contained py file that does all required imports, feeds all prompts through the added LLMs w the settings.

Doesn’t support Dalai.

[benwhalley]

I came here to ask about a similar request — specifically, whether it's possible to use a .cforge flow document from within code.

I guess what I was looking for was:

import chainforge as cf

forge = cf.load('example.cforge')
results = forge.process('mydata.csv')

results['prompt1']
# outputs a dictionary of results, one per line of the input

It would also be really cool if you could manipulate the graph from python, e.g.:

forge.prompts[1].models
# prints a list of current model slugs, e.g. 'ollama'

forge.prompts[1].models.append('gpt4') # set this prompt to also use gpt4
results2 = forge.process('mydata.csv')
# now includes gpt4 results

Would this be easy to implement, given the current server structure? I guess if the async llm calling needs a long running process then maybe this could be implemented as an api on the CF server that a python client (or any other lang) could call? e.g.:

forge = cf.forge('http://localhost:8000') # 
forge.open('example.cforge')
results = forge.process('mydata.csv')