Open-source sovereign AI framework
Build and deploy AI agents on infrastructure you control, using the models you choose. Suvereo provides the open-source foundation for connecting your data, systems and workflows, without locking you into a vendor or cloud provider.
The core
Chat, agentic executions, search, users and roles, storage and model choice ship with Suvereo, so you don't write any of it. It runs on your own servers or private cloud, inside your network.
The domain
The domain layer defines only what is specific to your field: its data, its integrations and how both reach agents. Enthusiast is the e-commerce one, with products and documents as data, and stores like Shopify and content sources like Sanity or WordPress as integrations. Synced data lands in your own database.
More on EnthusiastSide by side
We suggest one domain per Suvereo instance, with its own data, integrations and agents. A hospital could run one for clinical research next to one for operations, keeping research data and live patient data apart. They all share the same core, so an upgrade to the core applies to each of them.
The agents
Each agent takes on one job in your workflows, using your domain's data and tools, in a chat or as an agentic execution. A research agent could find similar cases across de-identified notes when a researcher asks. An operations agent could check tomorrow's schedule overnight and flag conflicts before the day starts. They use the models you choose. With a self-hosted model, prompts never leave your network.
Your domain and your agents are what you build, with your own team or with us. Everything else comes with Suvereo, and data leaves your network only where you send it: to a hosted model or a system you connect.
Conceptual: who is responsible for what, not services or data flow.
Two ways to run an agent, and everything around them. It works the same whatever the domain.
A person chats with an agent in the web UI, or in another system through the Suvereo API, such as a chatbot on your own site. Answers stream in, history is saved and files can be attached. Long chats are compacted so they stay manageable.
Another system starts a run through the API, with structured input, a structured result and no person in the loop. Inside the run, the output is checked and retried until it passes, and the caller sees finished or failed.
OpenAI, Azure OpenAI, Google, Mistral, Anthropic, or self-hosted models through Ollama, chosen per data set. With a self-hosted model, prompts never leave your network.
Deploy on your own servers or private cloud, with Docker Compose for a first install and Kubernetes for production. Your data stays in your systems, and Upside can build or run the setup with you.
Any number of MCP servers per data set, with OAuth sign-in. Their tools become available to your agents.
Semantic search over your domain's data, so agents find what is relevant by meaning, even when no word matches.
A separate workspace per department, research group or site, each with its own agents, integrations and models.
Custom roles, global or per data set, assigned to groups or straight to a user, and enforced on the API endpoints.
A full REST API with Swagger docs. Models, integrations and agents ship as packages; the core stays untouched.
Evaluation suites run your agents on test cases and write reports you can compare across models, prompts and code changes.
Start from what your agents will need: which data they search and which systems they act in.
The docs build an example domain file by file, with an agent on top.
Read the walkthroughE-commerce
Products and documents as data, seven store and content integrations, and six agents, from product search to invoice scanning. Suvereo was extracted from it, and Enthusiast 2.0 runs on it in its own repository.
Explore EnthusiastExample
What a clinical research domain could look like: de-identified clinical notes, a FHIR source that removes identifiers before anything is stored, and an agent that finds similar cases.
Your field
Start from your workflows and the systems they touch. Your team can build the domain, or we can build it with you.
Talk to us about your workflowsPython and Django on the back end, React in the UI, PostgreSQL for storage, and one Docker Compose file to start it. Your IT team can run and maintain it with skills it already has.
self-hostedgithub.com/upsidelab/suvereo · MIT licence
Read the docsWhat you can do with the code, and what Upside offers around it.
Licence
Use it, change it and ship it, commercially too. Keep the copyright and licence notice. The full text ships in the repository.
Hosting
It runs on your own infrastructure, operated by your team. Upside can set it up and provide managed services, so your team doesn’t carry it alone.
Support
Issues and questions go through the repository. For production use, Upside offers support agreements with agreed response times.
With Upside
We build domains and agents together with your team, from the first workflow to production.
Talk to us about your workflowsOnly if you choose a hosted model provider. With self-hosted models through Ollama, your data, prompts and results stay in your own environment.
Compliance comes from how a system is deployed and governed, and no framework provides it on its own. Suvereo gives you the controls: self-hosting, local models, and roles and permissions per data set. Upside can help design a setup that meets your requirements.
No. Enthusiast is one domain built on Suvereo, for e-commerce. Suvereo runs on its own, with a domain you write or with basic agents and no domain at all.
Three parts: entities, integration types and agent tools. The docs build an example domain file by file.
OpenAI, Azure OpenAI, Google, Mistral and Anthropic, or self-hosted models through Ollama. You choose per data set.
Yes. Another system can start an agentic execution through the API, with structured input and a structured result. Inside the run, the output is checked and retried until it passes.
Yes. One instance can run several domains, though we suggest one per instance to keep each system simple: instances side by side on the same core, each with its own data.
Yes. Evaluation suites run an agent on a set of test cases against real models, with mocked integrations, and score every run. You can compare prompts, models and code changes before a change goes live.
Upside maintains Suvereo and runs Enthusiast on it. Community questions and issues go through GitHub. For production use, Upside offers support agreements with service levels, and managed services in which we operate and maintain the setup on your infrastructure.
Want agents running on your own infrastructure? We build domains and agents together with your team.