Magnitude - Local AI Agent with Built-in Models

TL;DR

TL;DR: Magnitude is an open-source AI agent that bundles local models out of the box — no Ollama setup, no API keys, and no cloud dependency.

Source and Accuracy Notes

• Project page: magnitude.dev
• Source repository: github.com/magnitudedev/magnitude
• License: Apache 2.0 (verified via GitHub API license.spdx_id)
• HN launch thread: none found at time of writing
• Source last checked: 2026-08-17 (commit main, repo pushed 2026-08-17)

What Is Magnitude?

Magnitude describes itself as “an open source agent with local models built in.” Unlike Ollama (which requires separate model management) or Hermes (which is an agent framework that can use local models), Magnitude combines both: it profiles your hardware, downloads and configures compatible models, then runs the agent — all in a single install command.

From the README:

npm install -g @magnitudedev/cli
cd your-project
magnitude

The project runs on macOS and Linux natively, with Windows support through WSL. It uses a model catalog and hardware profiling to pick the right model for your machine.

Key Features

• Fully offline — prompts and files stay on your machine after download
• No API keys or token costs — models run locally via a bundled inference layer
• Hardware profiling — recommends the best model for your specific hardware
• Skills system — extend with Excel, PDF, Chrome, PowerPoint, Word, and other capabilities via a skills directory
• Open source — Apache 2.0 license, fully inspectable

Skills System

Magnitude ships a skills system for extending the agent. Skills are reusable capability packages installed via npx:

npx skills add vercel-labs/agent-browser   # drive your logged-in Chrome browser
npx skills add anthropics/skills/xlsx      # read and build Excel spreadsheets
npx skills add anthropics/skills/pptx      # build PowerPoint decks
npx skills add anthropics/skills/docx      # read and write Word documents
npx skills add anthropics/skills/pdf       # read, fill, and create PDFs

Skills are pulled from skills.sh, a directory service run by Vercel.

Privacy and Data Handling

According to the README, Magnitude does not send prompts or files to any cloud service. Once a model is downloaded, the agent runs entirely offline. This makes it distinct from cloud-hosted agents or API-based workflows.

You can also connect your own OpenAI-compatible inference endpoint if you prefer to route through your own server rather than the bundled local models.

Practical Evaluation Checklist

• Install: npm install -g @magnitudedev/cli — single command
• Run: magnitude — no daemon setup required
• Hardware: profiles automatically, no manual model selection needed
• Privacy: fully local, no cloud calls after initial download
• Extensibility: skills via npx skills add — no separate plugin framework
• License: Apache 2.0 — commercial use permitted
• Platform: macOS/Linux native, Windows via WSL

FAQ

Q: How is this different from Ollama? A: Ollama runs local models but is not itself an agent — you manage models and prompts separately. Magnitude bundles an agent framework with the model runtime in one package, so there is no separate Ollama installation or server to configure.

Q: Does it work without internet? A: Yes. After the initial install and model download, everything runs offline.

Q: Can I use my own model server? A: Yes. You can connect any OpenAI-compatible endpoint and use its models through Magnitude.

Q: What hardware is required? A: There is no fixed minimum. Magnitude profiles your hardware on first run and recommends the best model from its catalog for your configuration.

Q: Is this open source? A: Yes. The project is Apache 2.0 licensed and the source is on GitHub.

Conclusion

Magnitude is a turnkey local AI agent — it eliminates the multi-tool setup (Ollama + agent framework + model config) that usually blocks non-technical users from running local AI. The single npm install and magnitude command is the entire onboarding flow. If you want a private, offline-capable AI assistant on your machine without managing separate infrastructure, this covers the full stack in one install.