Start here

Getting Started Guide

This guide takes you from creating an Milly Lab account to your first useful result. It is for new users. It should take about ten minutes.

#Create an account

Go to https://m-lab.app/signup and choose one of:

• Sign up with email — enter your name, email address and a password. Milly Lab sends a verification email; open it and follow the link.
• Continue with Google — no password to manage. Milly Lab receives your email address, name and profile picture from Google.

If the verification email does not arrive, check your spam folder before requesting another.

#Find your way around

After signing in you land on the dashboard at /app. The main areas:

#Send your first message

1. Open Chat.
2. Pick a model from the selector.
3. Type your question and press Enter.

The response streams in as it is generated. You can attach files, use voice input, and turn on web search — those are covered in the Feature Guides.

Start on a small model. gpt-5.4-nano or deepseek-v4-flash are fast and cheap, and answer most questions perfectly well. Reach for a larger model when a small one has actually disappointed you, not before.

#Choosing a model

There is no single best model. The trade-off is speed and cost against depth.

Which of these you have depends on your plan. Plans are not cumulative — each includes its own specific set of models, so a higher plan is not a superset of a lower one. See "Plans, Billing & Usage".

If a model is greyed out, it is not in your plan.

#Understanding the allowance meter

Your plan gives you a monthly budget in US dollars, split across models, with each model having its own portion. So you can exhaust Claude while GPT still has plenty left.

Check what remains at Settings → Billing (/app/settings/billing).

Two things to know now, so they do not surprise you later:

• Cost grows as a conversation lengthens. The whole history is sent to the model each turn. Starting a new conversation for a new topic is cheaper as well as clearer.
• Swarm Mind costs much more than chat. One brief becomes many model calls. Use it for work that genuinely needs several models, not for questions a single chat could answer.

If you run out, Milly Lab refuses the request rather than billing you silently. The full list of payment-required errors is in "Plans, Billing & Usage".

#Try something bigger

Once chat feels comfortable:

Generate an image. Open Light Studio, describe what you want, and generate. Reference images and batch generation are there when you need them.

Run a Council. Open Council Mode, write one brief, and let several models answer it independently. Useful when you want to compare approaches rather than accept the first answer. You can diff the answers and merge the best parts.

Run a Project. Open Project Mode and give it a task with several parts. An orchestrator breaks it into subtasks and worker models execute them, with a synthesized deliverable at the end.

Swarm Mind requires a Max plan or above.

Connect a repository. In Project or Council Mode you can connect GitHub or Google Drive so runs read your actual files and propose edits. Nothing is ever written without your explicit approval — read the Connected Workspaces Guide before you connect anything.

#Set yourself up properly

Worth doing once, in Settings:

• Profile — set your interface language.
• Billing — see your plan and what is left.
• Team — if you are working with others, create a team and invite them.
• Agents — set per-model preferences for response length and reasoning effort.

#Getting a better result

• Say what you want, specifically. "Rewrite this paragraph for a non-technical reader, keeping it under 100 words" beats "improve this".
• Give the model the material. Attach the file or paste the text rather than describing it.
• Say what the output is for. A model writing for a board paper makes different choices than one writing a code comment.
• Check the output. Models produce confident text that is sometimes wrong. Verify facts, figures and citations before you rely on them.
• Leave web search off unless the answer depends on current information. It makes responses slower and costs more.

#Where to go next