CID

What you can do on STEAV

One platform for the whole life of a model: build it, test it, run it, improve it, and keep all of it. CID is the name of the platform; you will see it in the documentation.

Book a demo
Model registrylocal mesh · 6 modelsLocal
Models
deepseek-v4-flash
glm-5.2
kimi-k3
gemma-3-12b
clinical-notes-rag
fraud-detect-fed
Registry
Model registrySERVING
deepseek-v4-flash
VERSION4.0.1STAGEproductionDEVICEdgx-spark-01
Model registrySERVING
glm-5.2
VERSION5.2.0STAGEproductionDEVICEbyoc-b200 ×2
Model registrySTANDBY
kimi-k3
VERSION3.0.0STAGEstagingDEVICEbyoc-b200 ×12
Model registrySTANDBY
gemma-3-12b
VERSION1.0.2STAGEstagingDEVICErtx-5070-01
Model registryLOCAL
clinical-notes-rag
VERSION0.9.3STAGEdevelopmentDEVICEcpu-pool-02
Model registryLOCAL
fraud-detect-fed
VERSION3.2.0STAGEarchivedDEVICE
Sovereignty
HASHc3a1…8e07
RESIDENCYfail-closed
EGRESSallowlist only
RAGpgvector
SERVINGhardware you designate

Start from an open model. Make it yours.

Bring your documents, records, images, or transactions. The platform prepares the data and trains the model on it. You end up with a model that knows your business and belongs to you.

Train across teams without sharing the data.

Subsidiaries, partners, or sites that cannot pool their data can still train one model together. Each site keeps its data. Only the model updates travel.

Test before you ship.

A model cannot be released until it passes the checks you set. Accuracy, safety, and tampering checks run on every build, not just the first one.

Run it your way.

Self-hosted.

The platform runs entirely in your environment. Data and compute never leave.

Air-gapped serving.

Run the full AI inference stack with zero egress inside an isolated environment.

Hybrid.

Sensitive work stays in your environment. STEAV-operated capacity handles the rest.

STEAV-hosted.

We run the infrastructure for you, so your team gets the capability without the operational load.

Keep improving.

Retrain as your data changes. When a better base model comes out, move to it without rebuilding what you've already made.

See the history.

Every model keeps a record of the data, steps, and checks behind it. When someone asks how it was built, the answer is already there.

Manage everything from one place.

Models, tools, and agents under one dashboard, with the approvals and access rules you set.

Seven pipeline stages: data preparation, training, aggregation, evaluation, benchmarking, validation, deployment. Each stage signs what it did into one record.

Deploy on-prem, air-gapped, or in a region-pinned sovereign cloud with customer-controlled keys.

Drive pipelines from the terminal (CLI) or from an IDE and agent workflows (MCP server).

Also from STEAV

Amplify (coming soon).

Image and video models, built and run inside your own workflows.

Monolith (coming soon).

Labeling and annotation for the data your next model is built on.

Services.

Data, evaluation, and testing, done by our team when you want them.

See it with your own data.

Book a demo