This is a more advanced example, showing how to customize the way documents are processed and turned into memories when using the serverless memory synchronous pipeline.
The example uses memory.Orchestrator.AddHandler
to set which handlers are available:
memory.Orchestrator.AddHandler<TextExtractionHandler>("extract_text");
memory.Orchestrator.AddHandler<TextPartitioningHandler>("split_text_in_partitions");
memory.Orchestrator.AddHandler<GenerateEmbeddingsHandler>("generate_embeddings");
memory.Orchestrator.AddHandler<SummarizationHandler>("summarize");
memory.Orchestrator.AddHandler<SaveRecordsHandler>("save_memory_records");
and the steps
parameter to choose which handlers to use
during the ingestion:
await memory.ImportDocumentAsync(
new Document("inProcessTest")
.AddFile("file1-Wikipedia-Carbon.txt")
.AddFile("file2-Wikipedia-Moon.txt")
.AddFile("file3-lorem-ipsum.docx")
.AddFile("file4-KM-Readme.pdf")
.AddFile("file5-NASA-news.pdf")
.AddTag("testName", "example3"),
index: "user-id-1",
steps: new[]
{
"extract_text",
"split_text_in_partitions",
"generate_embeddings",
"save_memory_records"
});
Note that as soon as ImportDocumentAsync
is done, the memories are available for queries.
To run the example, either set the OPENAI_API_KEY
environment variable with your
OpenAI API key, or adjust the memory builder code to use Azure or other LLMs.