# Magnitude

> Frontier coding agent on open weights, in your infrastructure.

## Overview

Magnitude is a frontier coding agent, powered by open weights, that runs entirely in your VPC or data center. Your code never goes to Anthropic, OpenAI, or anyone else.

- **Product**: Frontier coding agent on open weights
- **Install**: `npm i -g @magnitudedev/cli`
- **Hosted app**: https://app.magnitude.dev
- **Backed by**: Y Combinator

## Performance

Running your own agent doesn't mean settling for a weaker one. On terminal-bench 2.1, Magnitude performs on par with Claude Code.

- Magnitude (GLM 5.2): 75.5%
- Claude Code (Opus 4.8): 78.9%

## Infrastructure

We deploy and tune the model serving stack inside your environment. Your code never goes to Anthropic, OpenAI, or anyone else.

- Developer's laptop runs the agent
- API gateway and GPUs run inside your environment (VPC or data center, optionally air-gapped)
- Your VPC or data center (optionally air-gapped)

## Controls

The governance and access controls enterprise teams run on, built in throughout.

- **SSO** — Single sign-on
- **RBAC** — Role-based access
- **Spend controls** — Budgets and limits
- **Audit logs** — Full activity trail

## Efficiency

Open weights cost a fraction of frontier-API pricing, so the same spend goes three times further. More PRs, more review, more shipped.

- Magnitude (GLM 5.2): 3x value ($142 total cost TB2.1 run)
- Claude Code (Opus 4.8): 1.0x baseline ($428 total cost TB2.1 run)

## Pricing

- **For individuals / small teams** — Runs on our infra, nothing to host. Zero data retention. $5 in free credits, then pass-through API pricing.
- **For enterprise** — Custom agents built around your stack. Run it anywhere: our infra, your VPC, bare metal, or air-gapped. SSO, RBAC, spend controls, audit logs.

## FAQ

- **Can I bring my own coding subscriptions or API keys?** No. We control the provider layer to ensure speed, reliability, and tool-call correctness.
- **How is Magnitude different from OpenCode?** OpenCode is a generalist harness. Magnitude is an opinionated harness, tuned specifically for open models. On Terminal-Bench 2.1, same GLM 5.2: we score 75.5%, they score 50.8%.
- **Can I host it on my own hardware?** Yes, if you are on team / enterprise plan.
- **Do you train on my prompts and code?** Absolutely not. Zero data retention, guaranteed.
- **Where are the models hosted?** When using our cloud, all models are hosted on US hardware, spread across geographic areas for low latency.

## Data & Privacy

- Zero data retention on prompts and code — guaranteed across the provider and all inference under the hood.
- No training on your prompts or code.
- All models hosted on US hardware, spread across geographic areas for low latency.
- External providers / BYOK not supported — Magnitude controls the provider layer to ensure speed, reliability, and tool-call correctness.

## Getting Started

```sh
npm i -g @magnitudedev/cli
magnitude
```

The CLI will ask for an API key. Sign up at https://app.magnitude.dev to get $5 in free starter credits.

## Links

- Website: https://magnitude.dev
- App: https://app.magnitude.dev
- Documentation: https://docs.magnitude.dev
- GitHub: https://github.com/magnitudedev
- Twitter/X: https://x.com/usemagnitude
- Contact: founders@magnitude.dev

## Keywords

coding agent, AI coding agent, open weights coding agent, terminal coding agent, coding CLI, open models, self-hosted coding agent, VPC coding agent, air-gapped coding agent, private coding agent, code review agent, AI developer tools, frontier coding agent