Models · Model agnostic
One platform. All models.
Choose between the latest models from the major providers, or run open source models on your own servers. Curiosity is not the model. It is what gives the model you choose its context.
Your model · your context
Three ways to run
Hosted, local, or your own
Which one fits depends on where your data is allowed to go.
- 01
Connect standard models
Ready to use models from the major commercial providers, with no setup beyond a key.
- 02
Run models on your system
Open weight models on your own servers, for full control and for estates where nothing may leave the building.
- 03
Build models for your needs
Models tuned to your data and your workflows, where the general ones do not know your vocabulary.
Models
The latest models, by where they run
27 current models from 6 providers, by where each one runs: a US server, an EU server, or on premises. 15 of them are open weights you can run on your own servers. Each name links to its provider's model card.
01 / 06
OpenAI
7 models
02 / 06
Anthropic
4 models
Claude Fable 5.1, model card, opens in a new tab
Yes on a US server No on an EU server No on premisesClaude Sonnet 5, model card, opens in a new tab
Yes on a US server No on an EU server No on premisesClaude Haiku 4.5, model card, opens in a new tab
Yes on a US server No on an EU server No on premises03 / 06
5 models
Gemini 3.8 Flash, model card, opens in a new tab
Yes on a US server No on an EU server No on premisesGemini 3.5 Flash-Lite, model card, opens in a new tab
Yes on a US server No on an EU server No on premisesGemma 4 26B A4B, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premises04 / 06
Mistral
6 models
Mistral Medium 3.5, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premisesMistral Large 3, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premisesMistral Small 4, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premisesMinistral 3 14B, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premisesMinistral 3 8B, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premisesMinistral 3 3B, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premises05 / 06
Meta
2 models
Llama 4 Maverick, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premisesLlama 4 Scout, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premises06 / 06
Microsoft
3 models
Phi-4-reasoning-plus, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premisesPhi-4-multimodal, model card, opens in a new tab
Yes on a US server Yes on an EU server Yes on premisesBuilt into Studio
19 providers, ready to connect
The tables above are the current models. Studio connects to each of these providers directly, and lists the models each one serves.
With your API key
OpenAI, Anthropic, Google Gemini, Mistral AI, xAI Grok, DeepSeek, Cohere, Perplexity, Moonshot AI, Z.AI, Alibaba Qwen, MiniMax, Upstage.
The model makers, each through its own API.
Inference platforms
Groq, DeepInfra, OpenRouter.
Open weights and other providers' models, served for you.
In your cloud account
Azure OpenAI, AWS Bedrock.
The models your cloud already offers, under your contract and in your region.
On your servers
Ollama, any OpenAI compatible endpoint.
Open weights on your own hardware. Nothing leaves the building.
Bring your own model
If it speaks a standard API, Studio can use it
Curiosity Studio is not tied to the list above. Point it at an OpenAI compatible endpoint, or write a provider for any other API, name the model, and the tools your teams build run on it.
OpenAI compatible
Any endpoint that speaks the OpenAI API. Most inference servers for open weights do, and so do the gateways platform teams put in front of their models.
Any other API
A provider you write in C# inside Studio, for an API that speaks neither. The example it ships with calls the Anthropic Messages API.
OpenRouter
Hundreds of models behind one key
OpenRouter, opens in a new tab is one of the providers built into Studio. One key reaches every model it lists, from many providers, through one endpoint. When a provider is down, OpenRouter falls back to another that serves the same model.
Use it to try a model the week it ships, before you sign with its provider. Requests to OpenRouter leave your environment. It stores no prompts unless you opt in, and it can route only to providers with zero data retention, opens in a new tab. For data that must stay in your building, run open weights on premises instead.
Why it is a setting
The model is the part that ages fastest
A system built around one provider has to be rebuilt when a better model appears, or when procurement changes its mind. Curiosity keeps the model behind an interface, so swapping it is configuration rather than a project.
The graph, the permissions and the tools your teams built on top do not notice.
Integrations
Integrations that work where you work
Curiosity connects to 70+ enterprise systems, so the model answers from your records rather than its training.
FAQ
Questions about models
Which model does Curiosity use?
The one you choose. 19 providers are built into Studio, among them OpenAI, Anthropic, Google, Mistral, Azure OpenAI, AWS Bedrock and Ollama for open weights on your own servers. Changing it is configuration. Curiosity is not the model. It is what gives the model context.
Can we bring a model that is not listed here?
Yes. Point Studio at any endpoint that speaks the OpenAI API, or write a provider of your own in C# inside Studio for any other API. That covers most inference servers for open weights, the gateways platform teams run, and the cloud endpoints the major providers offer in your region.
Can we use OpenRouter?
Yes. OpenRouter is one of the providers built into Studio, so one key reaches every model it lists, and you can limit which upstream providers it routes to. Requests to OpenRouter leave your environment, so keep it to data that may go to a hosted model.
Can we change the model later?
Yes. The model is a setting of the provider, so changing it is configuration. The graph, the permissions and the tools your teams built do not change with it.
Which models run on premises?
Open weight models, run on your own servers: 15 in the list above, from OpenAI, Google, Mistral, Meta and Microsoft. Nothing is sent to a provider. Ask us which fit the hardware you plan to deploy on.
Can I run models on my own system?
Yes. Run open weights with Ollama, or with any inference server that speaks the OpenAI API, on your own hardware. Studio calls it like any hosted provider.
Does the model see data a user cannot?
No. Permissions are resolved at query time, before the model is called, so Studio's built in tools only hand the model records the person asking may see. Tools you write in C# get a query scoped to that user too: use it.
Security
Built for security
Made in Germany and built for data sovereignty. Your data stays in your environment, encrypted and access controlled down to the individual record.
On premises or private cloud
Deploy on your own infrastructure or in certified data centers in Europe.
Single sign on and enforced MFA
Access synced to your data sources with relationship based permissions.
Full audit trail
Every query, download and data access logged.
Data residency
Your data stays in the region you choose.
Made in Germany
Built in Munich.
GDPR compliant
European data protection standards.
Your data. Your infrastructure.