bolt.new/packages/bolt/app/routes/api.enhancer.ts
2024-07-11 16:20:57 +02:00

66 lines
1.9 KiB
TypeScript

import { type ActionFunctionArgs } from '@remix-run/cloudflare';
import { StreamingTextResponse, convertToCoreMessages, parseStreamPart, streamText } from 'ai';
import { getAPIKey } from '~/lib/.server/llm/api-key';
import { getAnthropicModel } from '~/lib/.server/llm/model';
import { systemPrompt } from '~/lib/.server/llm/prompts';
import { stripIndents } from '~/utils/stripIndent';
const encoder = new TextEncoder();
const decoder = new TextDecoder();
export async function action({ context, request }: ActionFunctionArgs) {
const { message } = await request.json<{ message: string }>();
try {
const result = await streamText({
model: getAnthropicModel(getAPIKey(context.cloudflare.env)),
system: systemPrompt,
messages: convertToCoreMessages([
{
role: 'user',
content: stripIndents`
I want you to improve the following prompt.
IMPORTANT: Only respond with the improved prompt and nothing else!
<original_prompt>
${message}
</original_prompt>
`,
},
]),
});
if (import.meta.env.DEV) {
result.usage.then((usage) => {
console.log('Usage', usage);
});
}
const transformStream = new TransformStream({
transform(chunk, controller) {
const processedChunk = decoder
.decode(chunk)
.split('\n')
.filter((line) => line !== '')
.map(parseStreamPart)
.map((part) => part.value)
.join('');
controller.enqueue(encoder.encode(processedChunk));
},
});
const transformedStream = result.toAIStream().pipeThrough(transformStream);
return new StreamingTextResponse(transformedStream);
} catch (error) {
console.log(error);
throw new Response(null, {
status: 500,
statusText: 'Internal Server Error',
});
}
}