Help & FAQ

Three independent people from Switzerland, coming from different backgrounds. We built MAGUS because we were frustrated with bloated, corporate AI interfaces that feel like they're designed to upsell you rather than actually help you. We wanted something honest, capable, and built by people who actually use it every day. No VC money telling us what to ship. Just us and our users.

A few things that genuinely matter:

Multi-modal means you can chat, search the web, reason through hard problems, and generate images and video all in one place, switching modes per conversation depending on what you need.

RAG (retrieval-augmented generation) is available per agent, meaning you can give agents access to your own files and documents and they'll actually use that context when answering.

You can build custom agents, set up recurring jobs, use multiplayer conversations, and tap into a growing library of prompts, it's not just a chat window.

And honestly? We read every piece of feedback we get. Things that users ask for actually end up in the product. That's not marketing speak, that's just how small teams work.

We have a Discord community at discord.gg/6EfPDhmWRb, a support page if you want to reach us directly, and a blog where we share what we're building and thinking. Links are in the footer. Come say hi, we don't bite.

We integrate with multiple providers, OpenRouter, xAI, AIML API, and FAL for image and video generation, so you get access to a wide range of models without needing your own API keys or juggling multiple subscriptions. You pick the model, we handle the plumbing.

Credits reset at midnight in your local timezone every day. We track usage against your actual day — not some server timezone that's 9 hours off from where you live. You can set your timezone in account settings if it's not right.

Documents, plain text, images, video, and emails. Once uploaded, files get indexed so agents can search through them semantically — meaning they'll find the relevant section, not just dump the whole file into context.

Go to Settings > Preferences. There's a toggle for autoscroll there. Turn it off and you can scroll back freely while a response is still generating.

Rough guidance: for day-to-day chat and writing, a mid-tier model like Gemini Flash or Claude Haiku is fast and cheap. For complex reasoning or coding, go for Claude Sonnet, GPT-4o, or Grok. For image generation, check what's available in your image model selector. When in doubt, ask the AI itself it's surprisingly honest about its own limitations.

Just type "load skill onboarding" into any chat window and the AI will walk you through it.

Go to account settings. Currently we support English and German.

Two things to check: first, make sure you're in Image Generation mode, there's a mode selector in the chat input toolbar. Second, confirm you have an image model selected in your model settings. Both need to be set for image generation to work.

There's a memory system running in the background. After conversations, the AI extracts things worth remembering across three scopes: memories tied to a specific conversation, memories tied to your account that travel with you everywhere, and memories tied to a specific custom agent. You can turn global memory on or off in your account settings, and memories are searchable semantically so the right context gets pulled in rather than everything at once.

There's a dedicated Scheduled Jobs page in the main navigation. It shows all your jobs with their status, lets you filter by active or paused, and you can click into any job for the details.

The conversation owner's credits cover all AI responses in that conversation. If someone shares a conversation with you, their credits get used when the AI responds, not yours. Keep that in mind before handing out access to a busy conversation.

Your messages go to the model provider you've selected to generate the response, that's unavoidable, the model needs to see your message. What we control is our end: we don't train on your data, we don't sell it, and we don't log more than we need to. Each provider has their own retention policies, which you can find on their sites.

On servers in Europe. We're a Swiss company and that's a deliberate choice.

This is a production application built by a senior developer with real security practices in place, no exposed API keys, no leaky endpoints, encrypted in transit and at rest, standard access controls throughout. We take this seriously.

Settings > Subscription. You can see your current plan, your daily credit and storage usage, and available upgrades. If you have an active paid subscription, there's a Manage button that takes you into the Stripe billing portal for payment details and invoices.

Your subscription stays active until the end of the billing period you've already paid for. After that you drop to the free plan. Nothing gets deleted.

Your data stays as long as your account exists. Canceling a paid plan doesn't touch your conversations, files, or memories, you just lose access to the higher-tier features. If you want your account and data fully removed, reach out to us via the support page and we'll take care of it.