CocoIndex playground

Change a file. Follow the update.

You write the Python. CocoIndex keeps the index in sync.

Interactive demo · simulated

docs_pipeline / local example

Try a change

01 Source

3 files

refunds.mdeditable

Try editing the first text line.

02 Python flow

memo=True

parse()cached

Read the document text.

0 run · 3 reused

summarize()cached

Keep the first text line.

0 run · 3 reused

embed()cached

Create a vector from the summary.

0 run · 3 reused

Cached result = same function, same inputs.

03 Search index

doc_rows

Document / summaryLast sync

One row per file · summary + embedding

3 files. An index that is already fresh.

Your index is up to date. Change a file to see what runs again.

3/3 embeddings reused

Sample data. No API calls or setup.

flow.pysimplified example

1# Simplified flow; helper implementations omitted.2@coco.fn(memo=True)3async def parse(file): ...4 5@coco.fn(memo=True)6async def summarize(doc, *, detail="short"): ...7 8@coco.fn(memo=True)9async def embed(text): ...10 11async def process_file(file, target):12    doc = await parse(file)13    summary = await summarize(doc, detail="short")14    vector = await embed(summary)15    target.declare_row(row=DocRow(16        name=file.name,17        summary=summary,18        embedding=vector,19    ))

The browser simulates parsing, summaries, and embedding reuse. Your Python flow uses real functions and a real target.

Build this with your own data.Open the docs