Frequently asked questions

The public demo is a hosted dbsliceAI MCP server where there is only one database. All the tools are available to explore.

Yes, but not through the public demo. Private datasets need to be prepared in the dbsliceAI dataset format and served by a dbsliceAI-compatible MCP server. Contact ai@dbslice.org to discuss data preparation, deployment, and collaboration.

The dbsliceAI MCP server runs tools and returns selected outputs, such as summaries, plots, rendered images, or metadata, to the MCP client. Where those outputs go depends on the LLM that the MCP client is using. The LLM could be cloud-hosted, enterprise-secured, or fully-local. For private deployments, expose only the tool outputs you are comfortable sharing with that client.

dbsliceAI expects a structured dataset with item-level metadata (csv or json) and configured extracts. Extracts can include images (png), line data (csv or json), 3D surface data (glb), with optional embeddings. The exact preparation workflow depends on the source database and the analysis you want to expose.

No. The AI client requests operations using dbsliceAI's set of curated tools. Calculations, queries, plots, and rendering are performed by controlled routines.

No. CFD is the current example domain, but the underlying idea is broader: structured simulation or engineering datasets with metadata, cases, derived quantities, and visual extracts.

MCP provides a standard way to expose the tools to AI clients such as Claude.ai and ChatGPT while keeping operations explicit, inspectable, and repeatable.

dbslice is open source. dbsliceAI, including the MCP server and AI-assisted analysis workflow, is not currently open source.