Meet the new AI Magicx. We’re in closed beta. Public signup is closed.

Beta member sign-in

Chat · Auto · deep research

One chat. It picks the tools.

Leave Auto on and each message gets the model and the tools it needs. You see the estimate before anything that costs mana runs.

Live demo

One box. Seven tools.

Pick a request. The estimates are the real quotes the app would show.

Pick a request
Cycling through examples.
  • Web search (not used)
  • Deep research (not used)
  • Code sandbox (not used)
  • Artifacts (not used)
  • Images (used)
  • Video (not used)
  • Voice and audio (not used)

Request: Make a pixel sprite sheet of a mushroom hopping, three frames.

Routed toGPT Image 2.6 Flare
ToolsImages
Estimate, shown first≈ 7 mana

The tools

What each tool does.

  • Web search

    Searches the web and answers with numbered sources you can open. Up to 4 searches on an everyday message, 12 on a paid one.

  • Deep researchPro and Max

    Plans sub-questions, reads the best pages and writes a long, cited report. Shows the estimate before it starts.

  • Code sandbox

    Runs real code in a private sandbox: data analysis, charts, and files you can download.

  • Artifacts

    Documents, pages and small apps you can open, edit and download, kept next to the chat.

  • Images

    Makes and edits images with the Studio models, priced per image.

  • VideoPro and Max

    Short clips from a prompt or a first frame.

  • Voice

    Reads text aloud in the voice you pick.

Auto · On · Off

You stay in charge.

Every tool has three settings in the chat's Tools menu. Turn Auto off and only the tools you set to On are used.

  • Auto

    Magicx may use the tool when the message needs it. The default for every tool.

  • On

    The tool is used for this chat, even with Auto off.

  • Off

    Never used in this chat.

Auto picks the model too.

Everyday messages go to GPT-6 Luna. Long requests, big pieces of code and tasks that need careful reasoning can use GPT-6.1 Terra from Plus. Free stays on everyday models. Or pick a model your plan includes yourself and switch mid-chat.

Fast everyday models cost no mana up to your plan's daily chats: 20 a day on Free, 200 on Plus.

ROUTING SLIP


“Summarise this email”GPT-6 Luna
“Refactor these 400 lines”GPT-6.1 Terra
You picked a modelthat one

Deep research · Pro and Max

Long answers, every claim cited.

Claude Haiku 5.5 plans the questions and reads the pages; Claude Sonnet 5.5 writes the report. The estimate shows before it starts, and the hold is the most it can cost.

DepthSub-questionsSearchesPages readTypicalAt most
Standard51010≈ 17 mana39 mana
Thorough81616≈ 22 mana58 mana

The safeguard

Over 20 mana? It asks.

  • Any message estimated over 20 mana waits for your yes. Change the limit in Settings.
  • Mana is held when a job starts and charged for what actually comes back.
  • If nothing arrives, nothing is charged.
  • A schedule never approves anything itself: it pauses and emails you.

BEFORE IT RUNS


Deep research, thorough≈ 22
Most it can cost58

Over 20 mana, so Magicx asks you first.

Questions

About Auto.

What does Auto do?

Auto picks the model and the tools for each message. Everyday messages go to GPT-6 Luna; long or hard ones to GPT-6.1 Terra. With Auto on, a message can use web search, deep research, the code sandbox, artifacts, images, video or voice, and you can set each tool to Auto, On or Off. Anything estimated over 20 mana asks you first, and you can change that limit.

Is everyday chat free on paid plans?

Yes. Fast everyday models cost no mana, up to your plan's daily messages. After that, messages use a little mana.

Which models can I use?

59 models across chat, images, video, voice, music and transcription, including Claude, GPT, Gemini, FLUX, Kling and ElevenLabs. The models and settings you can use depend on your plan.

Choose how work runs

One task now. A batch for later.

Standard runs on every plan. Pro and Max also include Batch and Flex for text work. Model availability and the quoted price are shown before you start.

ModeBest forWhat to expect
StandardInteractive chats and tasks you need nowWork starts through the normal queue and replies arrive as it runs.
BatchMany independent promptsLower-cost processing on supported models, with results delivered later. Each item keeps its own progress and result.
FlexTasks with flexible timingLower-cost processing where offered, with variable capacity and delivery time.

The batch workspace keeps progress, results, cancellations and receipts together. Scheduled research and campaign copy can use available processing modes within the budget you choose.

Open the batch workspace