StellarCloud Overview
Managed open-source AI inference, EU-hosted, per-token billing. Every model StellarBase uses internally is available as a standalone API for your own applications.
What StellarCloud is
StellarCloud is a collection of production-grade AI models served over HTTP APIs. The same endpoints that power StellarBase internally are exposed for your own use — so you get the exact inference stack we trust for regulated work, without having to deploy it.
Key properties:
- Open-weights stack — leading open-source models for generation, embeddings, vision, and NER, plus our composite StellarOCR
- EU-hosted — Frankfurt, Amsterdam, Prague. No transatlantic data transfers.
- Zero retention — your prompts are not logged, not stored, not used to train anything
- OpenAI-compatible — drop-in replacement for apps that use the OpenAI SDK
- EUR billing — predictable per-token pricing, no hidden surcharges
Model families
Text processing
Language detection, lemmatization, named entity recognition, entity linking. Cheap, fast, multilingual. Detailed in Specialized Models.
Document processing
StellarOCR — one endpoint for text + layout + tables + figures + math from any PDF or image.
Embeddings & retrieval
Multilingual text embeddings (fast and high-recall tiers, plus a long-context variant), image embeddings, and rerankers.
Large language models
Open-weights LLMs for generation, reasoning, and coding — small, mid, and frontier tiers. Detailed in LLMs.
Smart routing
For the LLM endpoint, smart routing selects the cheapest / fastest / highest-quality model matching your constraints. You don’t have to pick a model by name unless you want to.
How it’s used
From StellarBase
Automatically. When your agent needs an embedding, or your workflow needs an LLM, StellarBase calls StellarCloud under the hood. No configuration required beyond the platform itself.
From your own apps
HTTP API, with a Python and TypeScript SDK. The chat endpoint is OpenAI-compatible (same request/response format), so existing OpenAI-based apps work by changing one URL.
Via StellarGate
If you want to use StellarCloud with PII anonymization — for GDPR-critical workloads — route your calls through StellarGate. Same API surface, privacy proxy in the middle.
Billing
Pay-as-you-go in EUR. Every API has a per-unit price (per million tokens, per thousand pages, per thousand images, per thousand searches). No minimums, no tiers — you pay for what you use.
See API pricing for the full breakdown.
Limits
| Limit | Default | Max on request |
|---|---|---|
| Requests per second | 50 | 500+ |
| Concurrent connections | 100 | 1,000+ |
| Max prompt length | 256K tokens | model-dependent |
| Max response length | 32K tokens | model-dependent |
| File upload (OCR) | 100 MB / file | 1 GB on request |
Reliability
99.7% SLA on the business-critical tier. Public status page. Incident history open. Multi-region failover for eligible endpoints. See Managed EU Cloud for operational details.
Self-hosted
The same models that power StellarCloud can run inside your own infrastructure. You pay an annual licence instead of per-token, get unlimited requests, and keep every byte in-network. See On-Premise.
Related
- Specialized Models — the full model catalog
- LLMs — generative models in detail
- Smart Routing — automatic model selection
- API Reference — endpoint-by-endpoint docs