The operations and governance layer for enterprise AI. Import models, serve them behind traffic-split endpoints, wrap them in guardrails, and hold the whole estate to EU AI Act controls — alongside the scheduling, experiment tracking and registry that turn a prototype into production.
Recorded from a live deployment — not a mockup.
Model Hub imports models straight from Hugging Face or your own registry. Models & Deploy puts them behind always-on endpoints with traffic splitting, alongside AI Studio flows and external providers like Azure OpenAI.
Guardrails Hub ships built-in detectors for PII, secrets and prompt injection, plus installable validators and your own registered guardrails — applied to every request, not bolted on per app.
Compliance tracks the EU AI Act controls in force — traceability, retention, human oversight, transparency, explainability, subject rights — with request logs, incident counts and a one-click audit export.
Import models directly from Hugging Face into your own environment, or register custom models. Each import lands in the AI Factory registry with status, modality and a ModelKit reference — the model comes to your infrastructure rather than your data going to the model.
Serve anything as an always-on endpoint: AI Studio flows, imported models, or external providers. Traffic-split groups let you run two versions side by side and shift load between them without touching consumers.
Built-in detectors cover PII, secrets and API keys, and prompt injection or jailbreak attempts. Add installed hub validators or register your own, so every served model inherits the same safety layer.
Controls in force are tracked article by article — Traceability (Art 12), Human Oversight (Art 14), Transparency (Art 50), Access Control (Art 15), Explainability (Art 13) and Subject Rights (Art 15/17) — with configurable log retention, incident tracking and audit export straight to your assessors.
Policies over who can call which model, enabled per user or group, evaluated on every request and surfaced alongside the compliance record.
Test any deployed model in the browser and copy the equivalent cURL. The API Guide walks key creation, base URL, chat completion and streaming, so an engineer is integrated in minutes.
MLflow-compatible run tracking, with versioning and alias-based promotion. Point production traffic at the production alias rather than a version number — repointing the alias is the promotion.
One scheduler for everything that runs on a schedule — AI Studio flows, Data Pipelines, notebooks and raw Spark jobs — with cron or plain-English schedules, retry intervals and failure counts.