Text, Streaming, and Tools

Use this reference for generateText, streamText, chat options, and tool loops.

Shared mental model

Most text APIs in xsAI use:

• apiKey
• baseURL
• model
• messages
• optional fetch, headers, abortSignal

For chat-style APIs, expect common options such as:

• temperature
• topP
• stop
• seed
• toolChoice
• tools
• maxSteps

generateText

Use generateText when the user wants one final result.

Returns a result shaped like:

• text
• finishReason
• usage
• messages
• steps
• toolCalls
• toolResults
• reasoningText

streamText

Use streamText when the user wants incremental output or tool events.

It returns immediately and exposes:

• textStream
• fullStream
• reasoningTextStream
• messages as a promise
• steps as a promise
• usage as a promise
• totalUsage as a promise

fullStream may contain events such as:

• text-delta
• reasoning-delta
• tool-call-streaming-start
• tool-call-delta
• tool-call
• tool-result
• finish
• error

Tool loops

• maxSteps enables repeated tool-use loops.
• Each step appends assistant output and tool results, then makes another API call if needed.
• Use toolChoice: 'required' when the model must call a tool.

Recommendation rules

• Prefer generateText for simple examples, scripts, and tests.
• Prefer streamText for UIs, live output, agent-like flows, and event-driven integrations.
• When the user wants a lightweight agent loop without a full agent framework, streamText plus tools is a good fit.