Try the API Talk to sales

Every layer of the model stack.

Train, tune, and deploy models with complete control. The Foundry hands you the weights, the runtime, and the keys.

Start building Enterprise rollouts
Offline mode inference 
from foundry import Runtime
from foundry.sampling import SampleSpec

model_id = "orrery/vega-small-2601"
sampling = SampleSpec(max_tokens=8192)

runtime = Runtime(
    model=model_id,
    tokenizer_mode="orrery",
    load_format="orrery",
    config_format="orrery",
)

messages = [
    {
        "role": "user",
        "content": "Who painted the Sainte-Colombe ceiling?",
    },
]

Model tooling, weights, and infrastructure. Deployable anywhere.

Own the full stack end to end. Fine-tune on your data, serve from your own hardware, and keep every weight, log, and key inside your perimeter.

Read the docs
Applied AI and deployment services
Model tooling
Agents Fine-tuning Embeddings Tool calling Distillation Guardrails Evals
Library of frontier models
Flagship language models Small and edge models Code models Multimodal models Bespoke models
Infrastructure management
Inference runtime Routing and caching Load balancing API gateway Resilience

Why the Foundry?

Complete model tooling.

Build durable AI products with deep customization, agent orchestration, and fine-tuning that runs on your own corpus.

The widest library of frontier weights.

Open and bespoke models across language, code, agents, and vision — with published benchmarks for each release.

Deployable anywhere.

Ship the same weights to edge devices, a private VPC, rented GPUs, or the rack in your own basement.

Applied engineering services.

Our residency engineers sit with your team from evaluation design through rollout, and stay through the first 90 days.

Own the next model you ship.

Train, tune, and deploy AI systems that stay entirely yours.

Try Sable Build on the Foundry Talk to an engineer