Foundation Models
Bring any model.
Keep your data.
Set the rules.
Connect the models you already pay for, or the one running inside your own network, and govern every call: who may use which model, where the keys live, what each request costs, and a record of all of it.
Bring your own model
Your model, wherever it runs
Point Scalata at any model endpoint: a model on your own servers, in your private cloud, or behind your internal gateway. It reads the models that endpoint serves, tests what each can do, and treats it like any other model in the workspace.
- Your data goes where your model is
When the model runs in your network, prompts and answers travel to your endpoint and nowhere else. No third-party AI provider sees them.
- Any model, any host
Open-weight models you serve yourself, models you have fine-tuned, a commercial provider we do not list, or a gateway your platform team already runs. If it serves a chat API, it connects.
- Tested before anyone uses it
Building the connection runs live tests against your endpoint and records what the model supports: streaming, tool calling, JSON output, images.
- Your costs, not ours
Set your own price per million tokens, so usage and cost reports reflect what the model really costs you to run.
Works with
- meta-llama/Llama-3.3-70B-Instruct
- Qwen/Qwen2.5-72B-Instruct
- mistralai/Mixtral-8x22B-Instruct
- deepseek-ai/DeepSeek-R1-Distill-70B
Providers
Or the providers you already use
Connect with your own account, under your own contract. Scalata finds the models your account can reach, so you add them by picking, not by typing.
AnthropicClaude models, with your own API key.
OpenAIGPT models, with your own API key.
Google GeminiGemini models, with a Google AI key.
Mistral AIMistral, Magistral and Codestral models.
Azure AI FoundryYour deployments, read from your endpoint.
AWS BedrockEvery model enabled in your account.
Google Vertex AIEnabled models, found across your regions.
Hugging FaceAny model with a live inference provider.
xAIGrok models, with your xAI key.
DeepSeekDeepSeek chat and reasoning models.
Fireworks AIOpen-weight models, served by Fireworks.
Security and access
Every model has an owner and a guest list
Only the people you allow can add a model, and every model carries its own permissions: who can use it, change its temperature, edit or delete it.
Grant access to a department and everyone in it has it. External users can be given use of a model through the chatbot, and nothing more. Keys are encrypted the moment they are saved and never shown again; colleagues can reuse a saved connection by its name without ever seeing the key. Anyone without access can ask for it, and the owner is told.
Tokens and cost
Every token accounted for
Each call is logged with its tokens, cost, speed and outcome, so you can see where the spend goes before the invoice does.
Break usage down by person and department, follow success rates and response times, and export it all to a spreadsheet. Monthly allowances and per-minute rate limits keep any one team or workflow from using more than its share.
Built for enterprise
Everything a model estate needs
Connecting a model is the easy part. Running dozens of them across teams, regions and vendors, safely and within budget, is what the rest of the platform is for.
If a model, provider or region is unavailable, requests move to the backup you chose, without anyone noticing.
Keep processing in the regions your policies and regulators allow, model by model.
Protect saved credentials with keys held in your own key management service, and revoke them there.
Reach the models you host over private connections, so traffic never crosses the public internet.
Redact personal data and block what your policy forbids before a prompt leaves, and check what comes back.
Spending caps per team, project or model, with alerts before they are reached rather than after.
Decide by policy which models each department, region or client may see at all.
Stream every call, with who made it and what it cost, into your SIEM or data warehouse.
Turn weeks of expert work into minutes
See how Scalata fits your team, your data, and your controls.