local-ai

MNN: Fast On-Device AI

Name: MNN: Fast On-Device AI
Author: alibaba

Get high-performance Edge AI with MNN, battle-tested by Alibaba, for startup founders

intermediate⏱ 30 minutes💵 Free (self-hosted)

16,018 stars2,429 forksC++Guide quality 9/10Updated 9/1/2026100% free · open source

What it is

Run highly optimized machine learning on small devices, like phones or robots, using MNN.

Real use case example

"A founder of a robotics startup wants to create a self-driving wheelchair. First, they connect a camera to their prototype. Then, they write code that uses MNN to detect obstacles and adapt the path. Finally, they test the device in different environments, refining its AI."

Best insideCursorClaude DesktopCodex CLI

When to use it

•When you need to deploy AI models on edge devices with limited computational resources
•When you want to improve the performance and efficiency of your on-device AI applications
•When you need a battle-tested and widely adopted inference engine for your startup's AI projects

Ready-to-paste prompt

mnn::Interpreter interpreter('model.mnn'); mnn::Session session(interpreter); session.run({'input': input_data}, {'output': output_data});

Heads up: MNN requires a compatible C++ compiler (e.g. GCC 5.4 or later) and a compatible OpenCV version (e.g. OpenCV 3.4 or later) to build and run successfully

Saves to your device

Use with Claude

New

Skip the builder — one click puts this in Claude, Cursor, Antigravity and more.

🛠️ Technical setup

Expect 20–40 minutes in a terminal — or let your AI agent drive it.

Try it instantly — no install

Claude Code

mkdir -p ~/.claude/skills/mnn && curl -fsSL https://workflowstacks.com/api/skills/mnn/claude-skill -o ~/.claude/skills/mnn/SKILL.md

Open in another AI app

Opens the app with this repo with the prompt ready to go — no copy-paste needed.

Connect the whole catalog (MCP)

claude mcp add --transport http workflowstacks https://workflowstacks.com/api/mcp

Adds a WorkflowStacks connector to Claude Code: search and load any skill here by chatting.

How MNN: Fast On-Device AI works

Codeflow

Free to inspect

MNN: Fast On-Device AI is a very large C++ project (~1.2M lines across 5457 code files, plus 572 test files). It is a full software project: use it through its install path rather than reading it end to end. Last commit this month, Apache-2.0 license, has a test suite.

Size

Very large codebase

~1.2M lines · 5457 code files · days to read — use, don't read

Setup

Developer setup

A real software project. Use it via its install path; don't expect to read it all.

Runs on

C++

No API keys detected

C++ 89%C 3%Python 2%Kotlin 2%Assembly 2%

Where to start reading

1
README.md
Start here — what it does and how to install it
2
AGENTS.md
The instructions the AI actually follows
3
source/backend/hexagon/README.md
Inside source/ — the main logic begins here
4
demo/exec/CMakeLists.txt
A worked example — copy this to get going

What's in each folder

source/Core code — the actual logic2641 files

skills/Prompts, skills & agent definitions53 files

apps/Sub-packages / plugins2733 files

docs/Documentation163 files

doc/Documentation154 files

demo/Examples you can copy18 files

3rd_party/Folder947 files

tools/Helper scripts842 files

READMEHas testsDocumentedCI checksExamples includedApache-2.0 licenseUpdated this month

Use with Claude

New

Skip the builder — one click puts this in Claude, Cursor, Antigravity and more.

🛠️ Technical setup

Expect 20–40 minutes in a terminal — or let your AI agent drive it.

Try it instantly — no install

Claude Code

mkdir -p ~/.claude/skills/mnn && curl -fsSL https://workflowstacks.com/api/skills/mnn/claude-skill -o ~/.claude/skills/mnn/SKILL.md

Open in another AI app

Opens the app with this repo with the prompt ready to go — no copy-paste needed.

Connect the whole catalog (MCP)

claude mcp add --transport http workflowstacks https://workflowstacks.com/api/mcp

Adds a WorkflowStacks connector to Claude Code: search and load any skill here by chatting.

Quick Actions

Details

Creator

alibaba

Language

C++

Related skills

More local-ai tools founders pair with this one.

local-ai★ 127,573

Run LLMs with llama.cpp

Get fast LLM inference with llama.cpp, for founders using C++.

local-ai★ 91,165

Stirling-PDF: Edit PDFs Anywhere

Edit PDFs on any device with Stirling-PDF, a tool for startup founders

local-ai★ 31,165

Run AI Locally with llmfit

Find compatible AI models for your hardware with llmfit, a tool.

local-ai★ 29,040

koreader: Multi-format eBook Reader

Get a versatile eBook reader for various devices.

local-ai★ 10,300

Run AI Locally with runanywhere-sdks

Run AI models on-device with this production-ready toolkit, ideal for startup founders

local-ai★ 9,622

amlogic-s9xxx-armbian: Run Armbian on Devices

Get Armbian on Amlogic, Allwinner, Rockchip devices with amlogic-s9xxx-armbian, for founders needing local AI solutions