Providers

Provider adapters normalize different AI streaming formats into store-ai's unified StreamEvent type. You choose a provider based on which API your backend streams from.

Built-in Providers

Adapter	Format
anthropic()	Anthropic SSE (message_start, content_block_delta, ...)
openai()	OpenAI Chat Completions SSE
openaiResponses()	OpenAI Responses API SSE
aiSdkDataStream()	Vercel AI SDK Data Stream protocol

Anthropic

import { createAIStore, anthropic } from '@store-ai/core';

const store = createAIStore({ provider: anthropic() });

Handles message_start, content_block_start, content_block_delta, content_block_stop, message_delta, and message_stop events. Supports text, thinking blocks, and tool calls.

OpenAI Chat Completions

import { createAIStore, openai } from '@store-ai/core';

const store = createAIStore({ provider: openai() });

Handles choices[0].delta.content, delta.tool_calls, and finish_reason fields.

OpenAI Responses API

import { createAIStore, openaiResponses } from '@store-ai/core';

const store = createAIStore({ provider: openaiResponses() });

Handles response.output_text.delta, function call argument deltas, and reasoning tokens.

Vercel AI SDK Data Stream

import { createAIStore, aiSdkDataStream } from '@store-ai/core';

const store = createAIStore({ provider: aiSdkDataStream() });

Consumes the Vercel AI SDK's UI Message Stream protocol. Useful when migrating from a Vercel AI SDK backend without changing the server.

Custom Streams

You can skip the provider layer entirely by passing a ReadableStream<StreamEvent> or AsyncIterator<StreamEvent>:

const store = createAIStore();

// ReadableStream
store.submit({ events: myCustomEventStream });

// AsyncIterator
store.submit({ events: myAsyncGenerator() });

See the ARCHITECTURE.md for the full StreamEvent type definition and how provider adapters are implemented.