Supercharge your AI agents with data from the web and beyond. Building the library for superintelligence. 🔥
TypeScript aiai-agentsai-crawlerai-scrapingai-searchcrawlerdata-extractionhtml-to-markdownllmmarkdownscraperscrapingweb-crawlerweb-dataweb-data-extractionweb-scraperweb-scrapingweb-searchwebscraping
Added instructions for empty string to extract prompts (#1114)
Rafael Miller948f7866df3c
diff --git a/apps/api/src/lib/extract/build-prompts.ts b/apps/api/src/lib/extract/build-prompts.tsindex 05f5bf6fb..6d89eb2f6 100644--- a/apps/api/src/lib/extract/build-prompts.ts+++ b/apps/api/src/lib/extract/build-prompts.ts@@ -97,7 +97,7 @@ export function buildBatchExtractSystemPrompt(): string {return ((systemPrompt ? `${systemPrompt}\n` : "") +- `Always prioritize using the provided content to answer the question. Do not make up an answer. Do not hallucinate. Be concise and follow the schema always if provided. If the document provided is not relevant to the prompt nor to the final user schema ${JSON.stringify(multiEntitySchema)}, return null. Here are the urls the user provided of which he wants to extract information from: ` ++ `Always prioritize using the provided content to answer the question. Do not make up an answer. Do not hallucinate. In case you can't find the information and the string is required, instead of 'N/A' or 'Not speficied', return an empty string: '', if it's not a string and you can't find the information, return null. Be concise and follow the schema always if provided. If the document provided is not relevant to the prompt nor to the final user schema ${JSON.stringify(multiEntitySchema)}, return null. Here are the urls the user provided of which he wants to extract information from: ` +links.join(", "));}diff --git a/apps/api/src/lib/extract/completions/singleAnswer.ts b/apps/api/src/lib/extract/completions/singleAnswer.tsindex 0f32d8c64..df66263af 100644--- a/apps/api/src/lib/extract/completions/singleAnswer.ts+++ b/apps/api/src/lib/extract/completions/singleAnswer.ts@@ -26,7 +26,7 @@ export async function singleAnswerCompletion({mode: "llm",systemPrompt:(systemPrompt ? `${systemPrompt}\n` : "") +- "Always prioritize using the provided content to answer the question. Do not make up an answer. Do not hallucinate. Return 'null' the property that you don't find the information. Be concise and follow the schema always if provided. Here are the urls the user provided of which he wants to extract information from: " ++ "Always prioritize using the provided content to answer the question. Do not make up an answer. Do not hallucinate. In case you can't find the information and the string is required, instead of 'N/A' or 'Not speficied', return an empty string: '', if it's not a string and you can't find the information, return null. Be concise and follow the schema always if provided. Here are the urls the user provided of which he wants to extract information from: " +links.join(", "),prompt: "Today is: " + new Date().toISOString() + "\n" + prompt,schema: rSchema,