Streaming Model

stream and streamSimple return an AssistantMessageEventStream, which is an async iterable of events. This stream is produced by provider adapters and is normalized to a single event model.

Event Types

AssistantMessageEvent in indusagi/src/ai/types.ts includes:

• start with a partial AssistantMessage.
• text_start, text_delta, text_end for text blocks.
• thinking_start, thinking_delta, thinking_end for reasoning blocks.
• toolcall_start, toolcall_delta, toolcall_end for tool calls.
• done with final message and stop reason.
• error with error message and stop reason.

Event Order

A typical streaming response:

• start
• zero or more block events in any order
• done or error

Tool call events can interleave with text or thinking blocks depending on provider. The final message can contain multiple blocks in message.content.

Result Handling

AssistantMessageEventStream implements result() which resolves to the final AssistantMessage. complete and completeSimple are simple wrappers that await result().

Tool Calls

Tool calls are represented as content blocks:

• ToolCall with id, name, and arguments.
• Arguments are parsed incrementally using parseStreamingJson.

Tool call IDs may be normalized to satisfy provider constraints. transformMessages in indusagi/src/ai/providers/transform-messages.ts:

• Normalizes tool call IDs for cross-provider replay.
• Removes unsupported signatures for other providers.
• Inserts synthetic tool results for orphaned tool calls.
• Drops assistant messages that ended with error or aborted.

Aborts

All streaming functions accept an AbortSignal in StreamOptions. Providers check signal.aborted and convert it to stopReason: "aborted".

Reasoning and Thinking

streamSimple converts SimpleStreamOptions into provider-specific options. It supports reasoning levels and optional thinkingBudgets. Provider adapters may clamp levels or convert them to provider-specific parameters.