Now onboarding early-access partners

Open-Source Model
Inference, Hosted

We're building a lean, no-nonsense inference API for open-weight language models. Same OpenAI-compatible protocol, competitive pricing, and direct access to the models that matter.

From Web Engineering to AI Infrastructure

Pixel Playroom LLC is a Florida-registered technology company founded by engineers with a decade of experience building and scaling production web systems. We've spent the last several years running Dockerized application stacks, managing nginx reverse proxies, and keeping e-commerce platforms alive under traffic — the same kind of plumbing that makes LLM inference work in production.

In 2024 we started shifting focus. The tooling around open-weight models had matured enough that running inference felt less like research and more like infrastructure — our territory. We began benchmarking vLLM, TensorRT-LLM, and SGLang, renting GPU nodes, and stress-testing serving configurations. By early 2025 we had a working OpenAI-compatible endpoint serving multiple models behind a single API key.

Today we're in private beta with a handful of design partners. Our stack is standard: vLLM on NVIDIA GPUs, Docker Compose for orchestration, Prometheus for monitoring. Nothing fancy — just what works.

Company

Legal Name Pixel Playroom LLC
Registered Florida, USA
Team Distributed (US / EU)
Stack vLLM · NVIDIA · Docker
API OpenAI-compatible
Status Private beta

What We're Serving

We don't try to host everything. We pick the models that actually get used and keep them fast.

DeepSeek V4 Flash
128K context
DeepSeek
Gemma 4
128K context
Google
Qwen 3.6 35B A3A
128K context
Alibaba
Ornich 1.0 35B
128K context
Ornich AI
Llama 4
256K context
Meta
Mistral Large
128K context
Mistral

Where We're Headed

We're building in public. Here's what we've done and what's next.

1

Infrastructure Setup

Benchmarked vLLM across A100 and H100 nodes. Established CI/CD pipeline for model deployment. Monitoring with Prometheus + Grafana.

2

Private Beta

OpenAI-compatible API live for a small group of design partners. Feedback loop active. Iterating on latency, throughput, and pricing model.

3

Model Marketplace Listing

Listing our models on aggregator platforms to make them available through existing tooling and familiar OpenAI-compatible interfaces. Expanding distribution alongside our direct API access.

4

General Availability

Public launch with credit-based billing, per-model pricing, usage dashboard, and self-service API key generation. Targeting Q4 2025.

Get in Touch

Interested in early access, partnership, or just want to talk infrastructure? We read every message.

Pixel Playroom LLC

We're based in Florida with team members across the US and Europe. Reach out and we'll get back to you within a business day.

📍
Florida, USA · Prague, Czech Republic
🏛
Pixel Playroom LLC