curated tools for accelerating

Lucidity.sh is an AI lab. We train compact model families that run on real hardware, and build the infrastructure and tools that put them to work.

Explore the models Read the research

Products

Where the models show up.

Ways to run our models and other people's, from raw API access to finished desktop tools.

Model lab

LucidityAI

The lab itself. Pico, Lumi, Synth, Pumori, Kasane, and Samsara are trained here. We focus on compact models for reasoning, STEM, creative writing, evaluation, and research that are small enough to run on hardware people actually have.

Model API

LuciditySH Platform

Direct, OpenAI-compatible access to Pumori S1, Synth 2.5, and Pumori Guard, metered per token with no monthly fee.

Inference API

Composite v11

An affordable, OpenAI-compatible inference API with free model access, subscriptions, routing, and usage accounting across upstream providers.

Roleplay studio

Stella

A creative workspace for AI characters, roleplay, and collaborative storytelling, powered by the Synth family.

Desktop agent

Vela

A lightweight agentic workstation for browsing, file work, commands, and coding, powered by Pumori.

How we train, and what we publish.

Research

Notes on creative writing evaluation, model behaviour, and what we learn from training small models. We publish them as we go, not as marketing.

Browse research →

Guardrails

Training priorities for models that augment people, respect privacy, preserve provenance, and resist surveillance or manipulation. Pumori Guard is trained against the same commitments.

Read the guardrails →

Build on models small enough to actually run.

Pull the weights, or point an existing OpenAI SDK at our API and start in one line.

Get the weights Use the API