What open data means when the software is yours to leave
The Curiosity Team
Industrial companies are careful about where their data lives. Aerospace, rail, energy and life sciences teams work under regulators, long asset lifecycles, and contracts that outlast most software. They have watched platforms arrive, take in their records, and make leaving expensive. So one question sits under every security review: if we put our knowledge here, can we get it back?
Much of enterprise AI answers that question badly. A system ingests documents, builds an index only it can read, and keeps the value in a format the vendor controls. The demo looks open; the exit does not.
Our position is that your company's data belongs to you, not to us. We want customers to keep Curiosity because it adds value, not because leaving is expensive. So we connect the records you already have, in the formats you already use, and you can always take them, and the structure we built around them, back out.
Structured formats in
Data comes in through standard formats and connectors, not a closed importer. Curiosity reads CSV, JSON, XML and Excel files, and extracts text from PDF and DOCX. It connects to SQL databases like PostgreSQL, MySQL and SQL Server, to cloud storage like S3, SharePoint and Google Drive, and to systems like Confluence, Jira and Salesforce. Most of these connect through Settings and Integrations; anything else, you build with the .NET connector SDK. Bringing data in still takes work, but through standard formats you control, not an importer only we control.
The same principle holds above the data. Identity runs on the standards you already use, OIDC, SAML, Entra ID, Okta and Auth0, with no proprietary identity layer in between.
Your graph, exportable in full
Every node, every property and every edge of the context graph exports as human-readable JSON, with no extra work on your side. Your schema, the definition of your node types and fields, exports as typed C# code you can compile against, so the model of your data lives in your own source control as well as in our interface. You can take full backups of the graph on your own schedule.
That is the concrete meaning of "not locked in": the format, the schema and the data are yours to take with you, at any time, with no export to negotiate.
Your infrastructure, your model
The same holds beyond data. Curiosity runs on your own infrastructure or a private cloud, deployed with Docker, Kubernetes, or an air gapped install, so your records stay in your environment. It connects the language model you choose: OpenAI, Azure OpenAI, Anthropic's Claude, or a local model served by Ollama, vLLM or LM Studio. Changing the model is configuration.
What to ask
If you are evaluating an enterprise AI system, ask how data gets in, how it reads back out, whose infrastructure it runs on, and whose model it depends on. Then decide whether the exit is as open as the demo.
To see how this works on your own data and infrastructure, talk to an engineer.
Read next
Articles on context graphs, enterprise search and industrial AI
