Generating text

generateText and streamText as Effects, streaming with Effect Streams, and how cancellation works.

generateText and streamText take the same options as the AI SDK versions. generateText returns an Effect of the same result; streamText returns an Effect of an Effect-typed stream result. The error type is AiError and the requirements are Ai plus whatever your callbacks use.

Callbacks are Effects

Pass an Effect wherever the SDK takes a callback. It can use any service available at the call site:

import { generateText } from 'ai-effect';
import { Effect } from 'effect';

const summary = generateText({
	model: 'openai/gpt-5',
	prompt: 'Summarize the water cycle.',
	onEnd: (event) => Effect.log(`Tokens used: ${event.usage.totalTokens}`),
});

Callback Effects cannot fail (their error type is never), so handle errors inside them. They run in the Ai layer, not in the fiber that made the call.

generateText

Interrupting the generateText Effect aborts the underlying request. An abortSignal you pass is still honored; either one stops the call.

streamText

streamText returns a StreamTextResult as soon as the stream is set up. It needs no Effect.scoped. Unlike generateText, it is not the SDK’s result: the streams are Effect Streams and the awaitable fields are Effects, all failing with AiError:

• Streams: textStream, partialOutputStream, elementStream and stream (the full parts, with error and abort parts emitted as data). The first three fail on an error or abort part instead of dropping it.
• Field Effects: text, content, usage, finishReason, toolCalls, toolResults, steps, output and the rest of the SDK’s promise fields.
• toUIMessageStream(), which converts the result for a chat UI; see UI message streams.
• raw, the SDK’s own result, for its response helpers such as toTextStreamResponse() and for consumeStream.

import { streamText } from 'ai-effect';
import { Console, Effect, Stream } from 'effect';

const chat = (prompt: string) =>
	Effect.gen(function* () {
		const result = yield* streamText({
			model: 'openai/gpt-5',
			prompt,
			onEnd: (event) =>
				Effect.log(`Streamed ${event.usage.totalTokens} tokens`),
		});
		yield* Stream.runForEach(result.textStream, (chunk) => Console.log(chunk));
		return yield* result.usage;
	});

Cancellation

Whether work stops depends on what you do with the result:

• Interrupting a fiber that consumes any of the streams, or one awaiting a field Effect such as text, aborts the call.
• Stopping early, for example with Stream.take, does not abort it. Neither does finishing or failing normally.
• Dropping the result without consuming it aborts nothing; the call runs to its end.
• Closing the Ai layer, or aborting an abortSignal you passed, still aborts the call.

Errors thrown while the stream is set up fail the streamText Effect itself. Errors during streaming fail the stream or field Effect with an AiError, both with method 'streamText'. See Errors.