Pearl Chat
v2.0
Pearl Chat
Getting started

Introduction

Welcome to the Pearl Chat v2 developer documentation

Pearl Chat provides open-weight native Luganda language models and streaming inference APIs built by the AI for Uganda Initiative. Our platform supports multi-turn conversational question answering, structured analytical reasoning, long-form story and poetry continuation, bidirectional Luganda to English translation, traditional proverb (Engero z'Abaganda) interpretation, and basic programming assistance in Luganda.

Get started

API quickstart Make your first streaming Pearl Chat API request in under 5 minutes. Complete API tutorial Walk through every task mode and endpoint with runnable Python examples. Conversational dialog and reasoning Hold multi-turn Luganda conversations and enable step-by-step thinking traces. Supported languages View supported languages, orthographic coverage, and the 12 Ugandan societal domains. Text and poetry continuation Complete Luganda sentences, stories, and long-form 500-word poems such as Kintu. Luganda basic coding Generate Python, HTML, CSS, JavaScript, and SQL code explained in Luganda.

Explore models

Pearl Chat models Run our 500M flagship and 60M to 350M Luganda instruction-tuned models efficiently on cloud GPUs or local consumer hardware.

Supported languages

Pearl Chat v2 is purpose-built for standard central Luganda (Oluganda olutuufu, ISO code lg / lug) and bidirectional English (en) workflows across 12 foundational sectors of Ugandan daily life.

For the complete breakdown of task modes, orthographic rules, and domain coverage, see our Supported languages reference.

Getting started

Ready to start building? Follow these steps to make your first request or load the weights locally.

1

Choose an endpoint or checkpoint

Connect to the hosted endpoint at https://pearl.aiforuganda.org/api/chat or download ai4ug/pearl-chat-500m-it from Hugging Face Hub.

2

Select a task mode

Specify dialog for conversational question answering, continuation for prose and poetry completion, or translate for Luganda and English translation.

3

Make a request

Stream tokens in real time over Server-Sent Events or run local quantized inference in Python.

Quickstart guide Get up and running in less than 5 minutes.

Community and support

Need help or want to contribute to sovereign Ugandan language intelligence?