Feature Request: a callback function with Chat and AsyncChat calls

[Karthik777]

both claudette and cosette have cb func and is useful for response manipulation. Would be great if lisette can support that.

I've currently patched it for my usecase with

@patch
def r(self:Chat,hid=-1,cb:callable=None,**kw):
    o=self.hist[hid].content
    return cb(o,**kw) if cb else o

[RensDimmendaal]

@Karthik777 thanks for creating the issue. We can probably add it to lisette too. Though tbh, I've personally never used the callback features in claudettte and cosette. Would you be able to give a concrete example of where it's useful for you? I'd love to learn.

[Karthik777]

Hey @RensDimmendaal thanks for reaching out. Completely missed this somehow.

I have a lisette chat as a tool like so:

class Step(BaseModel):
    explanation: str
    output: str

class PdfExtract(BaseModel):
    steps: list[Step]
    synthetic_data_json: str

def pdf_chat():
    pr='''
    Read through the claim form template and do the following:
    1 -> Generate a structured json object of all the fillable fields in the pdf form.
    2 -> create a mapping with potential value types for each fillable field. i.e. a yes and no question would have either "Yes" or "No" or understanding what the 4 options for a radio or a check field should be.
    2 -> Compare the structured json you've created with the extracted form fields along with potential options allowed passed below and combine them to create a final json object. MAKE SURE THE FINAL JSON OBJECT MATCHES THE KEYS FROM FORM FIELDS AND the POTENTIAL ALLOWED OBJECT VALUES AND MATCHES the STRUCTURE FROm THe FORM FIELDS. Make a judgment call which field keys match between the json object keys you create with that of the form fields.
    3 -> Generate synthetic data for the keys and create a final json object.
    return all the result in the structured data object attached.
    '''
    return Chat('gemini/gemini-2.5-flash',sp=pr)

def pdf_extract_tool(pdf_path: str # path to the pdf that needs to be extracted
) -> PdfExtract:
'''Extract the pdf in a structured format.'''
    pa = pdf_chat()
    rsp=pa(['Extract the pdf and generate synthetic data for:',Path(pdf_path).read_bytes(), jds(ext_fields(pdf_path))], response_format=PdfExtract)
    return jls(rj(pa.hist[-1].content))

Having a callback function is not necessary, but, I could have something like:

def ext(c): return jls(rj(c))

and in the pdf_extract_tool have something like:

return pdf_chat()(['Extract the pdf and generate synthetic data for:',Path(pdf_path).read_bytes(), jds(ext_fields(pdf_path))], response_format=PdfExtract, cb=ext)

now, I can have a string of tools each with callback functions to modify it to a specific format I need for the next chat.

Another use case is when I bind let's say a fn to store the history into sqlite as I go.

pa=pdf_chat()
chat_with_hst = bind(pa,cb=store2db)

Again, I can patch or add things as needed. But this is just a little easier.