Magnitude

What it does

Magnitude is an open-source inference engine for running open-weight models on a user’s own computer. When a model is downloaded, Magnitude tests kernel configurations on that hardware and keeps the fastest configuration it finds. It also includes speculative-decoding drafters, quantized KV memory, and shared prefix caches for concurrent sessions. The engine returns model inference locally; the desktop app lets users discover and download models, then connect an agent. A remote-server mode is also documented for serving browsers and agents on other computers.

Magnitude is intended for people who want to run local models or connect them to an existing agent, including users of Pi, OpenCode, Hermes, Codex, Claude Code, and Cline. Access is through a desktop app and CLI, with an OpenAI-compatible API for other agents. Install the app, download a supported model, and connect an agent. Supported systems are macOS, Linux, and Windows; hardware options listed include Apple Silicon, NVIDIA, AMD, Intel, and CPU-only machines. Model size depends on available memory and compute, and the supported model families are limited to the published list. The software is free under Apache 2.0; users provide the hardware and operating resources.

What it can do

• Runs open-weight models locally
• Tunes inference kernels for the user’s hardware when a model is downloaded
• Includes speculative-decoding drafters with models
• Uses quantized KV memory and shared prefix caches for agent sessions
• Connects to supported agents with a one-click flow
• Offers an OpenAI-compatible API for other agents

Works with

• Pi
• OpenCode
• Hermes
• OpenClaw
• Codex
• Claude Code
• Oh My Pi
• Cline

Before you choose

• Local model size and performance depend on available memory and compute; the publisher says there is no fixed minimum hardware requirement.
• Support is for the model families in Magnitude’s published model list, not all open-weight models.
• Remote-server mode is available on Linux and macOS; this is distinct from running the app locally on a desktop.

Listed under

• Coding
• Sandboxes and infrastructure
• Desktop app
• Developer tools
• LLM
• Local inference
• Open source
• Privacy

Get started

Install the desktop app and CLI, download a model, and connect an agent.

1. Install Magnitude using the macOS/Linux shell command, Windows PowerShell command, or a platform installer.
2. Open Magnitude, choose a recommended model in Discover, and download it.
3. Open Connections, select an agent, and follow the displayed command to connect it.

Free Apache 2.0 software with no token fee from Magnitude. Users provide hardware and operating resources.

Source: magnitude.dev, checked 1 Nov 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 1 Nov 2026. Each source supports part of the overview; availability and prices can change.

• Run open models as fast as your hardware allows | Magnitude ↗magnitude.dev
• Quick start - Magnitude ↗docs.magnitude.dev
• GitHub - magnitudedev/magnitude: Open source inference engine that tunes its kernels on your hardware to find the fastest way to run each model. 2x faster than Ollama. Works on Apple Silicon, NVIDIA, AMD, Intel, or just a CPU. · GitHub ↗github.com
• Introduction - Magnitude ↗docs.magnitude.dev

Company website

magnitude.dev: product, pricing and policy pages

Hacker News

Launch posts and discussion

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

manual_review

Last checked 1 Nov 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.