Quick Start Guide

Start here: Create an account, open chat, select a model and send a small task with a clear success check. Choose one model for the first answer rather than trying the entire catalog at once.

Updated October 28, 2026. Magicdoor combines chat, image tools, skills and usage tracking in one workspace. It is not a replacement for every feature of every provider's own app.

First chat

1. Open the model selector. Choose GPT-6 Luna, Claude Haiku 5.5 or GLM-5.3 Flash for a small first task, then compare the answer against your own requirements.
2. For manual Kimi work choose Kimi K3.1 Pro, not retired K3 or Renewable K3. Read the Kimi workflow and limits guide for migration, images, pricing and privacy boundaries.
3. Attach an image or PDF only when the selected model supports it. Pro accepts images but not PDFs or code execution; compatibility varies across the picker.
4. Enable Search on a supported model when the answer needs current sources. Magicdoor's web-search tool uses Exa; it is not an automatic handoff to a Perplexity answer model. Search can add separate charges.
5. Read the response, check any citations or proposed code, and review the recorded model and cost when available.

Experimental Auto chooses an answer model per message from a smaller pool; it does not choose freely from all 15 models. See Auto model selection. Astra, Opus and Fable require an active subscription. Browse model details for the complete current lineup.

Create or edit an image

Use the prompt-first Images workspace: select a model, add an image when editing is supported, describe the result and check the quote before sending. There is no need to find a separate Generate/Edit mode. You can also request image creation through a compatible chat model's tools.

Eight image tools include Google Nano Banana 2, Google Nano Banana Pro (2K), GPT Image 2.5 Flare, GPT Image 2.5 Sunburst, Seedream 5 Pro, Flux 2 Pro, Recraft V4.1 and Recraft Upscaler. Not every tool edits images: Recraft V4.1 generates; Recraft Upscaler increases resolution. GPT Image quality tiers cost $0.012–$0.50 per image; the upscaler is a separate $0.006 operation. Use the image model comparison for the full rate and input matrix.

Track your usage

Pricing explains Basic and Plus. Basic costs $6/month with $2 usage balance in month one and $1 on renewal; Plus costs $12/month with $7 monthly usage balance. Included balance resets each month; purchased top-ups do not expire. Your bill depends on tokens, selected models, tools and actual volume, not a guaranteed number of prompts.

Open Usage for balance, recorded usage and top-ups. Open Account Settings to manage your subscription. See model costs for token units, cache qualifications and context tiers.

No added consumer message-cap timer means neither unlimited capacity nor unrestricted model access. Balance, subscription-only models, provider outages and capacity, context limits and tool-specific controls still apply.

Preferences, memory and skills

Use chat preferences for defaults and instructions. Memory can provide reusable context; review what you save before including sensitive information. The memory migration guide explains how to transfer saved preferences deliberately. Skills supply reusable instructions for recurring work; they do not guarantee that every model supports the same tools or attachments.

Read the privacy policy for storage and provider boundaries. No-training routing is not end-to-end encryption or a zero-retention guarantee for your whole account.

FAQ

How much does Magicdoor cost?

Basic costs $6/month with $2 usage balance in month one and $1 on renewal. Plus costs $12/month with $7 monthly usage balance. Included monthly balance resets; purchased top-ups do not expire. Tokens and paid tools consume usage balance.

What models are available?

The picker has 15 chat models across eight inference operators. Use the model directory for the current lineup, capabilities and subscription requirements. Kimi K3.1 Pro replaces retired K3 and Renewable K3; eight image tools are available separately.

Does Magicdoor use my conversations to train models?

Magicdoor does not use your conversations to train models. Provider-inference policies and account/chat storage are separate; no training does not mean no retention. See the privacy policy and model-specific route notes before sending sensitive material.

How do I switch between models?

Open the model selector in chat and select a compatible model. Visible history can carry across, but attachment support, context limits and private reasoning differ by model. Review any compatibility warning before continuing.