<br />

> [!NOTE]
> **Note:** To add prebuilt vision, text, audio, or generative AI capabilities to your app in a few lines of code, start with [MediaPipe Tasks](https://developers.google.com/edge/mediapipe/solutions).

**MediaPipe Framework** is the low-level C++ and multi-platform graph engine
underpinning [MediaPipe Tasks](https://developers.google.com/edge/mediapipe/solutions). Use MediaPipe
Framework when you need to author custom C++ calculators, orchestrate
multi-stream timestamped packet graphs, or build bespoke GPU/CPU perception
pipelines.
![MediaPipe Framework CalculatorGraph architecture](https://developers.google.com/static/edge/mediapipe/images/solutions/diagrams/framework_graph.svg)

## Core concepts

Familiarize yourself with the three building blocks of a MediaPipe pipeline
before authoring custom graphs (see the
[Framework concepts overview](https://developers.google.com/edge/mediapipe/framework_concepts/overview)):
[Data Unit
### Packets
Immutable, timestamped data payloads (frames, tensors, landmarks, or protos) flowing along streams.](https://developers.google.com/edge/mediapipe/framework_concepts/packets) [Topology
### Graphs
Directed dataflow topologies connecting calculators and modular subgraphs using streams and side packets.](https://developers.google.com/edge/mediapipe/framework_concepts/graphs) [Compute Node
### Calculators
Modular C++ execution units implementing `Open()`, `Process()`, and `Close()` lifecycle hooks.](https://developers.google.com/edge/mediapipe/framework_concepts/calculators)

## Get started

Set up your build environment and compile example pipelines across desktop and
mobile targets:
[Setup
### Install MediaPipe Framework
Configure Bazelisk, OpenCV, and platform toolchains on Linux, macOS, or Windows.](https://developers.google.com/edge/mediapipe/getting_started/install) [Examples
### Build example applications
Compile and run MediaPipe Framework sample graphs for C++, Android, iOS, and Python.](https://developers.google.com/edge/mediapipe/getting_started/building_examples)

The open-source implementation is hosted in the
[MediaPipe GitHub repository](https://github.com/google-ai-edge/mediapipe), and
you can browse symbols in
[Google Open Source Code Search](https://cs.opensource.google/mediapipe/mediapipe).