ModelMessage

Defined in: packages/ai/src/types.ts:428

Type Parameters

TContent

TContent extends string | null | ContentPart[] = string | null | ContentPart[]

Properties

blockOrder?

optional blockOrder?: ModelMessageBlock[];

optional blockOrder?: ModelMessageBlock[];

Defined in: packages/ai/src/types.ts:452

The order of the blocks of an assistant message, when it is not the default order (all thinking, then the text, then the tool calls). Each entry points into thinking, into content (a string, by UTF-16 length), or into toolCalls. The library writes and reads this field. A reader that finds a map that does not match the message uses the default order.

content

content: TContent;

content: TContent;

Defined in: packages/ai/src/types.ts:435

createdAt?

optional createdAt?: Date;

optional createdAt?: Date;

Defined in: packages/ai/src/types.ts:482

Optional message creation timestamp. When present, message converters preserve it across persist → hydrate round-trips.

error?

optional error?: string;

optional error?: string;

Defined in: packages/ai/src/types.ts:454

Error reported by an AG-UI tool message.

id?

optional id?: string;

optional id?: string;

Defined in: packages/ai/src/types.ts:477

Optional stable message id. Providers ignore it; it exists so a persisted transcript can retain the streaming messageId and survive the persist → hydrate round-trip. When present, modelMessagesToUIMessages reuses it instead of generating a fresh id, so a hydrated message keeps the same identity as its live stream — which is what lets a mid-stream reload resume the SAME message bubble in place (see @tanstack/ai-persistence).

metadata?

optional metadata?: Record<string, any>;

optional metadata?: Record<string, any>;

Defined in: packages/ai/src/types.ts:462

Optional AG-UI message metadata. TanStack-owned fields live under tanstack.

midConversationChange?

optional midConversationChange?: MidConversationChange;

optional midConversationChange?: MidConversationChange;

Defined in: packages/ai/src/types.ts:460

The tools and system prompts that changed before this assistant message. chat() writes it when the adapter has a mid-conversation channel. Providers ignore it: adapters read TextOptions.midConversationChanges.

name?

optional name?: string;

optional name?: string;

Defined in: packages/ai/src/types.ts:436

role

role: "assistant" | "user" | "tool";

role: "assistant" | "user" | "tool";

Defined in: packages/ai/src/types.ts:434

structuredOutput?

optional structuredOutput?: StructuredOutputPart<unknown>;

optional structuredOutput?: StructuredOutputPart<unknown>;

Defined in: packages/ai/src/types.ts:468

Completed structured output represented by this assistant message. content remains the provider-facing JSON text; this field preserves the typed UI part across persistence and message conversion.

thinking?

optional thinking?: object[];

optional thinking?: object[];

Defined in: packages/ai/src/types.ts:444

Signed thinking to send back to the provider. redacted: true marks a block the provider encrypted: content is empty and signature holds its opaque data. See ThinkingPart.signature for the planned rename.

content

content: string;

content: string;

redacted?

optional redacted?: boolean;

optional redacted?: boolean;

signature?

optional signature?: string;

optional signature?: string;

toolCallId?

optional toolCallId?: string;

optional toolCallId?: string;

Defined in: packages/ai/src/types.ts:438

toolCalls?

optional toolCalls?: ToolCall<unknown>[];

optional toolCalls?: ToolCall<unknown>[];

Defined in: packages/ai/src/types.ts:437