Skip to main content
Tool calling (or function calling) is a language model’s ability to answer with a structured request to run a function: the tool’s name and its arguments in JSON, following a schema (usually JSON Schema) that the application defines. The model runs nothing itself: the application executes the function and returns the result so the model can continue.

​
Why it matters when generating documents with AI

  • It lets the model decide when a document is needed and request it with the right parameters (format, content, style).
  • The schema limits what the model can ask for, and the application can validate the arguments before running anything.
  • The result sent back to the model uses context: a short summary (status and link) is better than the whole file.
  • Errors returned as results, with a clear code, let the model fix its call.

​
With Doconda

doconda.tools() gives you the tools ready for function calling with Claude, OpenAI, Gemini, Mistral, Bedrock, Vertex AI or the AI SDK; they are JSON Schema, so they work as they are. Each tool returns a short summary with status and links, and errors come with a code the agent can understand (MCP server).