llmfn

Write Typescript / JavaScript function with natural language

Getting started

# npm 
npx jsr add @llmfn/llmfn

# deno 
deno add @llmfn/llmfn 
# or import directly via
import llmfn from "jsr:@llmfn/llmfn"

import llmfn, { setup } from "@llmfn/llmfn"

// set up configuration if you like to switch LLM Provider
// defaults to ollama 
setup({
  baseURL: "https://localhost:11434",
  apiKey: "ollama", // can be omitted if calling to ollama
  model: "llama3.3:70b", // specific model tag, defaults to "llama3:70b"
});

// 1. Define function signature
const fibonacchi = await llmfn<[number], number>({
  id: "fibonacchi", 
  inputs: [{
    type: "number",
    description: "number of iterations for the fibonacchi number",
  }],
  output: {
    type: "number",
    description: "the calculated fibonacchi number",
  },
})`calculate a fibonacchi number of given input iteration`;

// 2. Invoke function 
const result = fibonacchi(8);

// 3. Profit
console.log(`result of fibonacchi(8): ${result}`);

API References

llmfn()

Create tagged template function that takes in a string that describe the purpose of the function and let you generate JavaScript code using provided input JSON schemas, output JSON schema.

Parameters: signature (required) An object that defines the structure of the inputs, outputs, and the function ID.

properties:

id: An unique identifier for identifying the generated code, preferrably static in order to avoid generating the same function over and over again
inputs: An array of JSON Schemas defining the structure of each input argument.
output: A JSON Schema defining the structure of the function's output. Type: JSONSchemaType

Returns:

An asynchronous tagged template function which let you specify the prime objective of the function. What to do with the inputs and what you expect for the output.

setup()

Utility function for configurating LLM Provider

Parameters configuration (required)

An object that contains configuration you would like to update for your LLM Provider