An OpenAI-compatible API on GPUs in Greek data centres. Change one URL, keep your data in Greece.
The same inference core that runs our products, open to your applications: Greek-tuned open models, speech in and out, dedicated capacity, and the Company Brain runtime for your own cloud.
# Same SDK, Greek endpoint
from openai import OpenAI
client = OpenAI(
base_url="https://api.hellasinference.ai/v1",
api_key=os.environ["HI_API_KEY"],
)
r = client.chat.completions.create(
model="iris-8b",
messages=[{"role": "user",
"content": "Σύνοψη της κλήσης…"}],
)Endpoint and model names are illustrative; early-access customers receive theirs at onboarding.
The core under our products, as a service
OpenAI-compatible
Chat completions, streaming and tool calling through the API your code already speaks. Swap the base URL, keep the SDK.
Greek-tuned open models
Iris for instant answers, Metis for reasoning and tools, Clio for speech to text and Calliope for voice, routed per task. Each is built on an open-weight base with our own adapter. Model cards
Built for throughput
vLLM with FP8 and prefix caching on NVIDIA H200, tuned from our own benchmarks. By our estimate, ≈ 10 to 250 times cheaper than frontier APIs, depending on the task.
Dedicated capacity
Reserved GPUs in Greece for your workloads, from €50k per month, on take-or-pay terms you can mirror in your own contracts.
Company Brain runtime
License the runtime in your own cloud: ontology templates, a bitemporal fact store, a constraint gate where the model proposes and the rules decide, and an append-only evidence ledger.
Residency and compliance
Inference in Greek data centres, no US API in the path, and an AI Act (Regulation (EU) 2024/1689) dossier for every model we ship.
Built for teams that ship Greek AI
System integrators
Deliver Greek AI projects without sending client data to US APIs.
Software vendors
Embed Greek language and speech in your product on predictable infrastructure in Greece.
Developers
Prototype on shared endpoints and move to dedicated capacity when it matters.
Early access: the core is available to a limited number of partners while we bring dedicated capacity online. Tell us your workload and volumes.
Run your Greek workloads at home.
Tell us the models, volumes and latency you need. We will size it on our benchmarks.