Quickstart

This page indexes one document and searches it, on the speed profile: a 92 MB model, no reranker, no LLM and no API key.

From Python

import asyncio
from lasagna.sdk import KnowledgeClient, config_for_speed

async def main():
    client = KnowledgeClient(session_id="quickstart", config=config_for_speed())
    doc = await client.upload("paper.pdf")
    await client.documents.wait_for_processing(doc.document_id)  # upload returns once queued
    results = await client.search_documents("What did the ablation show?")
    for r in results.results:
        print(f"[{r.score:.2f}] {r.chunk.content[:120]}")
    await client.close()

asyncio.run(main())

upload() returns once the file is queued, so wait for processing before you search. wait_for_processing raises DocumentProcessingError with the reason and the fix if the file cannot be indexed.

Run it twice and the second run uploads nothing new: the document is reported with duplicate_of. Pass allow_duplicates=True to index it again.

From the shell

lasagna configure --profile speed
lasagna upload paper.pdf
lasagna search "what did the ablation show?"

lasagna init asks the same questions interactively. lasagna upload --force indexes a file again.

To serve the same index over HTTP, run:

lasagna serve

That starts the REST API and the configuration page on 127.0.0.1:8000. While it runs, the CLI asks the server instead of opening the store itself, because a local store opens in one process at a time.

What to expect

These timings are from an M1 Max (10 cores, 64 GB) on the speed profile:

Step	Time
First run, including the 92 MB model download	30 to 35 s
First run, model already cached	8 s
One search, one document (median of ten)	15 ms
One search over 18,442 chunks on Milvus Lite (median)	32 ms
One lasagna search command	about 10 s

A CLI command is slow because each one loads the model and opens the store. Use lasagna serve or the Python client when you search more than once.

The default profile (BGE-M3 plus the mxbai reranker) downloads about 3.2 GB and takes 117 ms per warm search on the M1 Max GPU and 3.1 s on two CPU threads.

How settings are kept

Settings belong to a session. Each session has its own embedding model, and its data lives in ~/.lasagna. A client created without config= reads the session's saved settings. See Configuration for the rules about which setting wins.

Next

Your first comparison tests a reranker of your own.