Integrate Screenpipe for AI-powered Data Retrieval
LlamaIndex reader for Screenpipe: indexes local OCR screen captures and audio transcriptions for RAG search.
Maintainer of this project? Claim this page to edit the listing.
0.14.23Add to Favorites
Why it matters
Leverage your local Screenpipe recordings (screen captures and audio) to build powerful AI applications. This integration allows you to easily index and query your personal data for enhanced research and summarization.
Outcomes
What it gets done
Connect to a local Screenpipe instance.
Load screen capture (OCR) and audio transcription data.
Filter data by content type, query, and time range.
Index and query your personal recordings using LlamaIndex.
Install
Add it to your toolbox
Run in your project directory:
curl -fsSL https://spark.entire.vc/get/li-reader-readers-screenpipe | bash Overview
LlamaIndex Readers Integration: Screenpipe
A LlamaIndex reader for Screenpipe, loading locally recorded screen OCR text and audio transcriptions as documents filterable by content type, query, and time window, ready for vector-indexed search. Use when building a personal RAG or search tool over screen and audio activity already recorded by a running local Screenpipe instance.
What it does
This skill is a LlamaIndex reader integration that pulls data from a locally running Screenpipe instance - a 24/7 screen-and-microphone recording tool that captures screen content via OCR and audio via Whisper transcription, storing everything in a local SQLite database behind a REST API. The ScreenpipeReader connects to that local API (default http://localhost:3030) and loads captured content as LlamaIndex documents, ready to be indexed and queried like any other data source.
Its load_data() method supports filtering by content type ("all", or narrowed to just audio transcriptions), a text query filter (e.g. searching for "meeting notes"), and a time window (e.g. only the last 24 hours via start_time=datetime.now() - timedelta(hours=24)), so a user can pull a targeted slice of their screen/audio history rather than the entire local recording archive. Once loaded, the documents plug directly into a standard VectorStoreIndex and query engine, letting a user ask natural-language questions against their own recorded screen and audio activity - the source's own example asks "What did I discuss in my last meeting?" against loaded audio transcriptions.
When to use - and when NOT to
Use this skill when building a personal knowledge base or RAG application over what a user has seen and said on their own machine - recalling meeting discussions, retrieving what was on screen at a certain time, or searching personal activity history - and when a local Screenpipe instance is already running and recording.
It is not useful without Screenpipe actually installed and running locally first, since the reader is purely a client for that local REST API and does no recording or OCR/transcription itself. It's also inherently scoped to what Screenpipe has already captured - it can't retrieve activity from before Screenpipe was recording or from a different, unrecorded machine.
Inputs and outputs
Input is a content-type filter (all or audio-only), an optional text query, and an optional time window, passed to ScreenpipeReader().load_data(). Output is a list of LlamaIndex Document objects containing the matching OCR screen text and/or audio transcriptions, ready to build into a VectorStoreIndex for semantic search and question-answering.
pip install llama-index-readers-screenpipe
Integrations
Requires a local Screenpipe instance (github.com/mediar-ai/screenpipe) exposing its REST API, and integrates directly with LlamaIndex's VectorStoreIndex and query engine for building searchable indexes over the loaded documents.
Who it's for
Developers building personal RAG or search tools over their own recorded screen and audio activity via Screenpipe, who want that data queryable through LlamaIndex rather than only browsable through Screenpipe's own interface.
Source README
LlamaIndex Readers Integration: Screenpipe
Reads screen capture (OCR) and audio transcription data from a local
Screenpipe instance.
Screenpipe is a 24/7 local screen & mic recording tool that captures screen
content via OCR and audio transcriptions via Whisper, storing everything in a
local SQLite database and exposing a REST API.
Installation
pip install llama-index-readers-screenpipe
Usage
Make sure Screenpipe is running locally (default: http://localhost:3030).
from llama_index.readers.screenpipe import ScreenpipeReader
from llama_index.core import VectorStoreIndex
reader = ScreenpipeReader()
### Load recent screen and audio data
documents = reader.load_data(content_type="all", limit=50)
### Load only audio transcriptions with a query filter
from datetime import datetime, timedelta
documents = reader.load_data(
content_type="audio",
query="meeting notes",
start_time=datetime.now() - timedelta(hours=24),
)
### Build an index and query
index = VectorStoreIndex.from_documents(documents)
engine = index.as_query_engine()
response = engine.query("What did I discuss in my last meeting?")
FAQ
Common questions
Discussion
Questions & comments · 0
Sign In Sign in to leave a comment.