Skip to content

Inspecting calls

See exactly what was sent, what came back, what it cost, and what would be sent, without sending it.

import functai
functai.configure(lm="gpt-4.1-mini", temperature=0)   # the model behind every output on this page
from functai import ai, _ai

When an answer surprises you, look at the conversation before changing anything. Nearly every problem is visible there: an instruction that says something else than you meant, an input that arrived empty, a reply in the wrong form.

@ai
def summarize(text: str, focus: str = "key points") -> str:
    """Summarize the text in one sentence, concentrating on the focus."""
    ...

summarize("FunctAI lets developers write typed functions whose body is a model call, "
          "so they can concentrate on logic instead of prompt strings.", focus="benefits")

'FunctAI benefits developers by allowing them to write typed functions with model calls as the body, enabling them to focus on logic rather than crafting prompt strings.'

The last calls: phistory

print(functai.phistory())

[2026-10-02T09:57:59] summarize → gpt-4.1-mini

System message:

Function: summarize

Summarize the text in one sentence, concentrating on the focus.

Reply in exactly this form:
<result>
...
</result>


User message:

<text>
FunctAI lets developers write typed functions whose body is a model call, so they can concentrate on logic instead of prompt strings.
</text>
<focus>
benefits
</focus>


Response:

<result>
FunctAI benefits developers by allowing them to write typed functions with model calls as the body, enabling them to focus on logic rather than crafting prompt strings.
</result>

(finish: stop; tokens in 83, out 38)

phistory(3) shows the last three calls. functai.inspect_history(3) returns them as objects (lm15's Request and Response), to inspect in code.

Before sending: render

fn.render(...) builds the exact request the call would send, without sending it:

request = summarize.render("Some text.")
print(request.system)

Function: summarize

Summarize the text in one sentence, concentrating on the focus.

Reply in exactly this form:
<result>
...
</result>

The layout: explain and signature_text

functai.signature_text(summarize)

'Signature: summarize | Doc: Function: summarize\n\nSummarize the text in one sentence, concentrating on the focus. | Inputs: text:str, focus:str | Outputs: result*'

print(summarize.explain())

adapter: functai_xml
reader: derived
input  text                 kernel-scalar (kernel)
input  focus                kernel-scalar (kernel)
output result               kernel-scalar (kernel)

What a call cost

fn.predict(...) returns the tokens, summed over every model call it made (tool loops included):

p = summarize.predict("Short text.")
p.usage

{'input_tokens': 60, 'output_tokens': 14, 'total_tokens': 74, 'cache_read_tokens': 0, 'cache_write_tokens': 0, 'reasoning_tokens': 0}

For a whole run, evaluate and fn.map put the tokens and time of every row in their table; see Evaluation.

One line per call

functai.configure(debug=True) prints a line for each call as it happens: which function, which model, how long, how many tokens.