Last updated:
AI Studio
AI Studio puts AI models to work inside Any2Info: answering chat questions, extracting data from documents, or running as a step in a dataflow. Setting up an agent follows a fixed path: pick a provider, connect it, build the agent, and optionally expose your own dataflows back out through MCP.
Agents
Section titled “Agents”An agent is configured with a provider and model, a system instruction that defines its role and boundaries, typed input and output parameters, and sources it can draw on: dataclips (what exists), dataflows (what it’s allowed to do), files, and vector stores. It can also call external MCP tools. History and file-retention settings control how much of a conversation is remembered and for how long.
Providers
Section titled “Providers”Before anything else, you need an account with a supported AI provider. Any2Info currently supports OpenAI, Azure AI, Claude, and Mistral, each requiring its own account, project or resource, and API key, set up directly with the provider.
Connections
Section titled “Connections”A connection is how Any2Info reaches a provider: created in Data Studio as an API headers connection, with the actual API key set separately afterwards via Set values so secrets stay out of the connection definition itself. One connection can be reused by multiple agents.
Create agent
Section titled “Create agent”Building an agent means configuring its model (direct model selection for OpenAI, Claude, and Mistral, a resource and deployment name for Azure AI), its instructions, its input/output parameters, and the sources and external MCP tools it may call. An agent can’t return files directly: to produce a document, have it return structured text and trigger a dedicated dataflow.
Where agents consume external tools, MCP lets Any2Info expose its own dataflows and dataclips as an MCP server for external AI clients such as OpenAI to call. Each configuration lists which dataflows and dataclips are active capabilities, and the generated endpoint is secured with OAuth 2.0.