README

sidebar_key

llm-api-objects

sidebar_category

text-content

LLM API Objects

Typed PHP objects for Chat Completions requests (messages, tools, tool calls) that work with both OpenAI and Ollama. Works with both OpenAI and Ollama (via the /v1 OpenAI-compatible endpoint) using the openai-php/client.

Why

A chat completion request is a deeply nested array: roles, content, tools with JSON-schema parameters, tool calls and their ids. Built by hand, a typo becomes an API error at runtime, and a long document has to be cut to fit the context by you.

LLM API Objects builds the request from typed classes (Chat, Message, Tool, ToolFunction, ToolParameter), so the IDE and static analysis catch the mistakes before the call.

OpenAI and Ollama with the same objects, through Ollama's OpenAI-compatible endpoint.
Long content split for you. A message is split into chunks by a delimiter and by an approximate token limit (maxTokens).

Features

Strongly-typed models for Chat Completions requests
Full support for function calling (tools): Tool, ToolFunction, ToolParameter, ToolChoice
Multi-turn conversations with role=tool messages and tool_call_id
Automatic message splitting when content exceeds a token limit
Delimiter-based message partitioning
Compatible with OpenAI and Ollama (/v1 endpoint)
Ollama model tuning parameters via ModelFileParameters
Jira XML/RSS importers to seed conversation context

Documentation

Getting Started — Installation and first request
Chat Model — Building and sending chat requests
Messages — Roles, content, tool_call_id, tool_calls
Tools — Function calling: Tool, ToolFunction, ToolParameter, ToolChoice
Model Parameters — Ollama model tuning parameters
Importers — Jira XML/RSS importers

Quick Examples

Basic chat with OpenAI

<?php
require 'vendor/autoload.php';

$client = OpenAI::factory()
    ->withApiKey($_ENV['OPENAI_API_KEY'])
    ->make();

$chat = new \ByJG\LlmApiObjects\Model\Chat(
    model: 'gpt-4o',
    messages: [
        new \ByJG\LlmApiObjects\Model\Message(
            role: \ByJG\LlmApiObjects\Enum\Role::User,
            message: 'What is the capital of France?',
        ),
    ],
    system: 'You are a helpful geography assistant.',
);

$result = $client->chat()->create($chat->toApi());
echo $result->choices[0]->message->content;

Basic chat with Ollama

<?php
$client = OpenAI::factory()
    ->withApiKey('ollama')
    ->withBaseUri('http://localhost:11434/v1')
    ->make();

$chat = new \ByJG\LlmApiObjects\Model\Chat(
    model: 'llama3',
    messages: [
        new \ByJG\LlmApiObjects\Model\Message(
            role: \ByJG\LlmApiObjects\Enum\Role::User,
            message: 'Hello!',
        ),
    ],
);

$result = $client->chat()->create($chat->toApi());
echo $result->choices[0]->message->content;

Function calling (tools)

<?php
use ByJG\LlmApiObjects\Enum\ToolChoice;
use ByJG\LlmApiObjects\Model\Chat;
use ByJG\LlmApiObjects\Model\Message;
use ByJG\LlmApiObjects\Enum\Role;
use ByJG\LlmApiObjects\Tool\Tool;
use ByJG\LlmApiObjects\Tool\ToolFunction;
use ByJG\LlmApiObjects\Tool\ToolParameter;

$weatherTool = new Tool(
    new ToolFunction(
        name: 'get_current_weather',
        description: 'Get the current weather in a given location',
        parameters: [
            new ToolParameter(name: 'location', type: 'string', description: 'City name', required: true),
            new ToolParameter(name: 'unit', type: 'string', enum: ['celsius', 'fahrenheit']),
        ],
    ),
);

$chat = new Chat(
    model: 'gpt-4o',
    messages: [
        new Message(role: Role::User, message: "What's the weather in Paris?"),
    ],
);
$chat->addTool($weatherTool);
$chat->setToolChoice(ToolChoice::Auto);

$result = $client->chat()->create($chat->toApi());

Installation

composer require byjg/llm-api-objects

Running Tests

composer test

Open source ByJG

byjg / llm-api-objects

Package info

Security

Statistics