Open Magi is the open agent runtime that enforces your rules: define the checks your work needs, and the runtime enforces them before every response ships.

What is Open Magi Cloud?

Open Magi Cloud is the hosted version of the open agent runtime. It adds managed accounts, encrypted secrets, isolated workspaces, model credits, monitoring, and support.

Can I self-host Open Magi?

Yes. Open Magi is open source first. You can self-host the runtime, bring your own model keys, and use Open Magi Cloud when you want managed hosting.

What AI models are supported?

Open Magi is designed for Claude, GPT, Gemini, local models, and future providers. The work agent layer should not be tied to one model vendor.

Do I need to operate infrastructure?

No. Developers can self-host, but Open Magi Cloud handles hosting, accounts, billing, secrets, storage, monitoring, and upgrades for teams that want a managed service.

Yes. End-to-end encryption for API keys and bot tokens (AES-256-GCM), isolated Kubernetes containers per user, data stored in EU. Compliant with GDPR, CCPA, and Korea's PIPA.

How much does it cost?

The open source runtime is free to self-host. Open Magi Cloud plans start with Pro at $14.99/mo and include managed hosting plus LLM credits billed at provider cost plus VAT only.

What can my AI agent do?

Open Magi can retrieve team context, draft from source material, call specialist tools, compare files and decisions, prepare briefings, monitor changes, verify outputs, and save useful reasoning or artifacts back into memory for the next task.

Can I use my own API key?

Yes. The BYOK plan lets you connect your own Anthropic, OpenAI, or Google API key. You pay API costs directly to the provider; Open Magi handles hosting, memory, skills, and everything else.

Tailored to youFollows your rulesOpen source

The AI agent runtime you can actually govern.

Name: Open Magi
Price: 7.99 USD
Author: Open Magi

Wrap any model in executable policy, evidence, and approval gates. It follows your rules instead of just hearing them.

Already on Claude Code? Add a control plane

Won't make things upStays in its laneShows its work

ClaudeGPTGeminiLocal modelsYour tools

Magi agent· #client-comms

policy on

Email the Q3 numbers to the client.

Drafting the email and checking it against your policy.

Policy check

VERIFIEDevery figure traced to source

BLOCKEDexternal send · rule: external-sends-require-signoff

HOLDINGwaiting for your approval

sealed to ledger #4821 · append-only

Reply “approve” to sign off, or revise.

Plain English in. A machine contract out, enforced by code, not a prompt.

Sound familiar?

Generic agents don't know your rules. So they break them.

They look great in a demo, then cite a number they never checked or skip the step that mattered. Magi follows the rules you set, on every task, and stops when it can't.

How it works

Three steps. About three minutes.

Tell it your rules.

In plain English. No code.

“Every claim must cite a source.”

“Reconcile every number to its source.”

“No external message without my sign-off.”

“Stay within the $10k spend limit.”

It builds your agent.

Around your work, tools, and limits.

Allowed read your sources, check citations

Blocked running code, deleting files

Required a source behind every claim

Ask first before sending anything out

It checks every action.

Against your rules, on every task.

Verified every source checked

Caught a number that didn’t match its source

Fixed corrected before it reached you

Paused waiting for your OK to send

Why Magi

Built to be trusted, not just smart.

Won't make things up

Before it answers, it checks its work against your sources. No source, no claim.

Stays in its lane

It only does what you allow, inside the limits you set. Over the line, it stops and asks.

Shows its work

Every step is right there. See it, audit it, undo it.

Build locally

Run it yourself. Keep everything.

Magi is open source. Run the agent yourself, bring your own model keys, and keep your data where your work already lives.

Run the agent yourself
Bring your own model keys
Inspect everything it does
Keep data in your own environment

Use Cloud

Or let us host the boring parts.

Want the agent without running servers? Cloud handles accounts, secrets, storage, and support, with LLM credits at 0% markup.

0% LLM markup. You pay provider cost plus VAT. That's it.

Hosted workspaces and storage
Encrypted secrets and isolated runtimes
0% LLM markup: provider cost plus VAT
Monitoring, upgrades, and support

Open source plus hosted

Start with source. Upgrade to hosted.

Hosted accounts, encrypted secrets, managed runtime, and LLM credits billed at provider cost plus VAT only.

Self-host

Open Source

Free

Run the core runtime on your own infrastructure.

Open runtime
Bring your own keys
Local or server deploy

View source

Hosted

Cloud Pro

$14.99/mo

Managed hosting plus LLM credits billed at provider cost.

Credits: $5/mo LLM credits
Search: 500 Brave searches/mo
Knowledge Base: 5GB Knowledge Base
Runtime: 16GB bot storage

0% LLM markup
No API key required
E2EE secrets

More credits

Cloud Pro+

$89.99/mo

More at-cost inference budget and support for heavier work.

Credits: $80/mo LLM credits
Search: 1,000 Brave searches/mo
Knowledge Base: 50GB Knowledge Base
Runtime: Priority support

Everything in Pro
0% LLM markup
Expanded Knowledge Base

Dedicated

Cloud Max

$399/mo

Dedicated runtime capacity with larger at-cost LLM credits.

Credits: $350/mo LLM credits
Search: 4,800 Brave searches/mo
Knowledge Base: 500GB Knowledge Base
Runtime: Dedicated node, up to 5 bots

16GB RAM
140GB dedicated storage
QMD Vector Search
0% LLM markup
Discord integration

FAQ

Questions, answered.

What is Magi?

Magi lets you build an AI agent tailored to your work, one that follows the rules you set. You describe how your work should be done in plain English, and Magi builds an agent that checks every action against those rules before it acts.

Really 3 minutes?

Yes. The hard part (verification, permissions, and logging) is already built in. You just add your rules, so getting a working agent takes minutes, not days.

Do I need to code?

No. You describe your rules in plain English. Developers can go deeper with the open-source runtime, but it's not required.

How is this different from OpenClaw, Hermes, and other agents?

Most off-the-shelf agents do the same thing for everyone, and you just hope they follow your instructions. Magi builds an agent around your specific rules and enforces them on every action, so it can't quietly skip the step that matters.

Can I self-host it?

Yes. Magi is open source first. Self-host the runtime, bring your own keys, and switch to Cloud whenever you want managed infrastructure.

Which models does it support?

Claude, GPT, Gemini, and local models. Magi is provider-neutral, so you can swap models behind the same rules.

openmagi.ai

When no one's watching, your agent still follows your rules.

Run it under your policy, with proof. Get started in 3 minutes.