Models, runtimes, and evaluation research — engineered to hold in production.

OpceanAI builds AI models, container infrastructure, and evaluation research — engineered under real constraints.

10+ models

One lineage, YuuKi v0.1 to RxG, with six specialized branches.

4 isolation layers

Doki runs OCI containers natively on Android.

289 models

Moud, our platform, in Public Beta. One catalog across providers.

Since 2023

Open evaluation research — NHE and the Imprint Theory.

The model ledger

One lineage and its branches, maintained as a working catalog.

Public Beta · Now open

Moud

Models, APIs, agents, compute, deployments, and developer tools in one place.

After a few days of development, testing, breaking things, fixing them, and probably breaking them again, Moud is entering Public Beta. What started as a fairly simple idea grew into an AI and developer infrastructure platform built by OpceanAI.

289

Models reachable through a single platform: language and reasoning models plus other AI capabilities from different providers, and a selection we host ourselves on Moud infrastructure.

The catalog keeps changing as models, providers, and infrastructure availability change.

More than a model playground. You can build on top of the catalog with:

  • API access
  • BYOK
  • Agents
  • Compute
  • Virtual machines
  • Deployments
  • Managed DNS
  • Moud subdomains
  • Channels and integrations
  • Developer tools

You can also register a device and sign in without copying an API key every time.

Privacy

Moud is built with privacy and data minimization in mind. Selected services hosted directly by Moud operate with Zero Data Retention. Third-party models remain subject to their own providers’ policies.

Infrastructure

Operations run across Mexico, Canada, and the United States, with Moud’s primary operations based in Mexico.

How this is funded

A mix of our own infrastructure, provider capacity, shared resources, and usage limits keeps a significant part of the platform available for free. One person’s money is another person’s tokens.

There will be bugs. Models may disappear. Providers will probably have problems at some point. That is exactly why the beta is open: use Moud, find the rough edges, and tell us what to build next.

doki · termux — arm64

$ doki pull alpine:3.20

pulling manifest sha256:beefdbd8 … done

extracting 3 layers … done

$ doki run -d --name sync-agent alpine:3.20

9f3c1ab2

$ doki ps

CONTAINER IMAGE STATUS ISOLATION

9f3c1ab2 alpine:3.20 Up 12s L2 · proot

41d7e09c nginx:1.27 Up 3h L3 · ns

c208aa41 redis:7.4 Up 26m L2 · proot

$ doki stats sync-agent

CPU 2.1% MEM 18.4 MiB NET 0.6 KiB/s

$ ▌

Doki

OCI containers, natively on Android.

Standard Docker images run unmodified on-device — no cloud, no remote daemon. Four isolation layers, from sandboxed to system-level, selected per environment.

In development

ASL — Android Subsystem for Linux. A complete Linux kernel in userspace, built in C++ and Rust. Kernel-first — work begins after Doki v1.

57.6%

Average token reduction, measured. Tsuki — trained on 4,160 bilingual examples across six task types.

NHE — Not Humanity Exam

Existing benchmarks measure what a model knows. NHE measures how human it still thinks — six cognitive patterns structurally embedded in human language, traced inside artificial systems. It is the empirical implementation of the Imprint Theory.

Latest

Platform

Moud enters Public Beta: 289 models, APIs, agents, compute, deployments, and developer tools in one platform.

mound.opceanai.com
Program

OpceanAI for startups is open: domains, compute, storage, and our engineers for open-source projects and early teams.

startups.opceanai.com
Support

OpceanAI is supporting a startup for the first time. We are providing Badtheorylabs with compute capacity to help bring Tinfield 1 to production; their API is live at rntm.sh.

rntm.sh

Evaluating the models, running Doki, or talking research — reach out directly.

Contact us