A function is the contract between the application and the AI model. It defines the desired input and output.

We specifically separate out functions and their defintions because in machine learning, its often common to swap out model parameters, what model architecture is used, or even if you’re using an LLM vs a Heuristic vs a combination of both to get the answer.

The actual implementation (prompt or other settings) of the function is described in an impl.

With baml, you can modify the implementation of a function and keep the application logic that uses the function unchanged.


inputAny type or with named args:
(<name>: <type>, ...)
The input type of the functionYES
outputAny typeThe output type of the functionYES
default_implstringThe name of the impl to use when the application logic doesn’t specifyOnly if you define multiple impls for a function. See impl

Examples - How to call functions

See the “generated python code” tab to see how to call the function in python.

Single Argument with Primitive Type

function GetSentiment {
    // In the case of a single argument, you can specify the
    // type directly.
    input   string[]
    output  string
from baml_client import baml as b

# GetSentiment is a function defined in the baml file
# response is a string
response = await b.GetSentiment(["I love this movie", "it is amazing"])

Class argument

class CustomerInfo {
    name: string
    age: int
    address: string

function SummarizeCustomerDetails {
    // In the case of a single argument, you can specify the
    // type directly.
    input   CustomerInfo
    output  string
from baml_client import baml as b
# Import the CustomerInfo pydantic model we generated
from baml_client.baml_types import CustomerInfo

response = await b.SummarizeCustomerDetails(
    CustomerInfo(name="John", age=30, address="123 Main St"))

Named-arguments (multiple arguments)

enum Color {

function GetSentiment {
    // You can have as many arguments (1+) as you want, but
    // they must be named if specified between parentheses.
    input   (query: string, millis: int)
    output  Color[]
from baml_client import baml as b

# Since GetSentiment has multiple arguments, each
# argument is passed in as a named argument.
# response is List[Color]
response = await b.GetSentiment(query="I love this movie", millis=1000)