PureBrain Onboarding Guide

Everything you need from first login to full partnership -- in one place.

AI Provider Account

PureBrain is your AI operating team. The compute comes from a separate AI provider -- currently Claude by Anthropic. You need an active Claude account connected to PureBrain before anything else works.

How it works
Think of it like hiring a team. PureBrain is the team. Your AI provider account is the office space where they work. You need both.

Service Connections

Your AI starts as a chatbox. These connections turn it into a real assistant that can email, manage files, and publish web pages. Each service gets its own account.

Don't worry about the technical steps
Your AI will walk you through each setup. Things like "app passwords" or "access approvals" might sound intimidating -- your AI will explain exactly what to click.

Start with these two

Separate Gmail -- For Real-World Communication Do First

A new Gmail account just for your AI, so it can email people and access Google services.

Sign up: accounts.google.com/signup -- Format: youragentname.makr@gmail.com

"Help me create and set up your Gmail account. Walk me through each step."

AgentMail -- Your AI's Email Address Do Second

Gives your AI its own email to send and receive messages on your behalf.

Sign up: agentmail.to -- Format: youragentname@agentmail.to

"Set up an AgentMail address for yourself, then send me a test email."

Add when ready

Google Drive & Calendar Recommended

Drive = shared filing cabinet. Calendar = schedule visibility. Both use your AI's Gmail -- no separate signup.

"Walk me through sharing my Google Drive and Calendar with you."

Cloudflare -- Free Web Publishing Recommended

Your AI can create professional web pages with shareable links -- dashboards, reports, anything.

Sign up: cloudflare.com/sign-up (free)

"Walk me through setting up Cloudflare so you can publish web pages."

Telegram
Your AI sets up a Telegram bot during its initial awakening. No extra setup needed. If you don't use Telegram, the portal works great on mobile too.
About passwords
Your AI stores passwords in a secure file, not in the chat. After any restart, ask: "Check that all your logins still work."

Pre-Day-One Checklist

Ready to Launch
Infrastructure in place. Time to meet your team.

Day 1: Context First, Tasks Second

Your first session is not about tasks. It is about context. The more your AI understands about you, the better every future interaction becomes. Think of it as onboarding a new hire.

Share: Your role and industry. Your clients or key stakeholders. How you prefer to communicate. Your quality standards. Key terms and jargon.

The Critical Instruction

After sharing your context, say: "Save everything I told you to permanent memory." Without this, it forgets everything when the session ends.

Good vs. Bad Prompts

Specificity is the difference between useful output and generic filler.

Vague
Help me with my proposal.
Specific
Research CompanyX's last 3 quarterly results and draft a one-page competitive brief for my Thursday pitch to ClientY.
Vague
Write me a blog post.
Specific
Write a 600-word post on [topic] in my voice -- direct, no fluff, data-backed. Reference the article I published last month on [X].
Vague
What is blockchain?
Specific
My client asked about blockchain in supply chain. Prepare a 5-bullet brief for tomorrow's call, focused on logistics companies under $50M revenue.

Pick One Workflow and Delegate It

Not a question. Not a brainstorm. A real task with a real deliverable.

Consultant

"Prepare a pre-call brief for my meeting with [client] on Thursday. Include their recent news, our last 3 interactions, and 3 talking points."

Coach

"Review my client's progress notes from the last 4 sessions and draft a summary with patterns and 2 suggested focus areas."

Freelancer

"Draft a proposal for [project] based on the scope we discussed. Use my standard pricing. Keep it under 2 pages."

Agency Owner

"Compile this week's content calendar: 3 LinkedIn posts, 1 blog outline, and 2 client social posts. Use the brand voice guide."

Tier-specific guidance

Awakening (1 AI): Focus on one workflow. Depth over breadth.

Partnered (2 AIs): Define roles -- one for client-facing work, one for internal operations.

Professional (3 AIs): Assign departments -- client work, research, content/biz dev. Clear lanes.

🚀
First Delegation
You stopped asking and started delegating. That is the shift.

Two Types of Memory

Understanding the difference is the single biggest factor in whether PureBrain compounds in value or stays flat.

Working Memory
The whiteboard.
Everything in the current conversation. Has a size limit. When full, it gets wiped ("compacted"). This is the default.
Permanent Memory
The filing cabinet.
Information you explicitly tell your AI to save. No limit. Survives across sessions and restarts. You have to ask for it.

The CTX Meter

The bar at the top of your screen shows how full working memory is. It changes color as it fills:

0-60%
Healthy
60-80%
Getting full
80-95%
Compact soon
95%+
Critical

When the whiteboard fills up, it gets wiped -- but the filing cabinet stays. This is called "compacting" and it's completely normal. Your AI will usually do it automatically, or you can type /compact in chat.

What to Save (Exact Phrases)

"Remember that my client ABC Corp prefers formal language and their fiscal year ends in March."
"Save this correction: I want bullet points, not paragraphs, for client briefs."
"Remember this decision: we are going with Strategy A because [reason]."
The Rule

Every time you correct your AI, end with: "Save this so you don't repeat the mistake."

Pre-Compact Checklist

Copy and paste this to your AI before any compact:

"Before you compact: (1) save this conversation as a document, (2) save any unfinished work to a file, (3) update your memory with everything important from this session -- decisions, preferences, corrections, (4) write a handoff note covering what we did, what's pending, and what to do next."
What survives vs. what doesn't
Keeps: Saved memories, preferences, files, passwords, connected services.
Gone: Current conversation, anything discussed but not saved.

Session Handoffs

At the end of every session:

"Write a handoff summary covering what we did, what is pending, what you need from me, and decisions I made. Save it to memory."

At the start of every session:

"Check your memory and the last handoff before we start."
HANDOFF -- April 27
COMPLETED: - Drafted proposal for ClientX (saved to /proposals/clientx-april.md) - Researched 3 competitors for Thursday pitch PENDING: - Waiting on ClientX budget confirmation before finalizing scope - Blog post outline needs your review NEXT SESSION: - Finalize proposal if budget confirmed - Review and edit blog outline - Prepare Friday team meeting agenda DECISIONS MADE: - Going with Option A pricing (higher margin, fewer deliverables) - Using case study format for the pitch deck
Time investment
5 minutes at the end of a session saves 30 minutes at the start of the next one.

Privacy Boundaries

Your AI is a professional partner, but there are boundaries. Do not share:

  • Client passwords or login credentials
  • Bank account details or financial account numbers
  • Sensitive personal data of third parties (health records, legal matters)
  • Information you would not put in an email to a trusted assistant

The Feedback Loop

  • Good output: Say what was good and why. "This brief was exactly right -- the tone matched the client, and the bullet format made it scannable."
  • Bad output: Say what was wrong, why, and tell it to save. "Too formal for this client. Conversational tone for internal updates. Save this."
  • Decisions: Save the reasoning. "We chose vendor B for API reliability, not price. Remember this."

Portal Customization

The portal has animated 3D backgrounds that can drain battery on tablets and laptops. Ask your AI to simplify:

"Use simpler backgrounds and turn off the 3D animations."

You can also switch dark/light mode, change colors, or adjust layout. Tell your AI and it will remember.

Important
When customizing, your AI should only change visual styling -- not underlying server code. If it suggests changing anything beyond look and feel, say no and contact support.

Scheduled Tasks (BOOPs)

A BOOP is a recurring task your AI runs on a schedule -- check email, scan for news, prepare a daily brief. Think of it as a standing instruction: "do this every hour" or "do this every morning." Your AI keeps working even when you are not in the portal.

Most useful recurring task: email monitoring. Ask your AI to check your inbox on a regular cycle -- flag what is urgent, draft responses where appropriate, and summarize what came in. This means you never miss an important email even when you are busy.

"Check my email every hour. Flag anything urgent immediately. Draft responses for routine items and hold them for my review."

Other recurring tasks: overnight research, daily news scans, weekly report preparation, document processing.

Always include a safety timer:

"For scheduled tasks, set a 60-minute max. If stuck for 5 minutes with no progress, shut down automatically."
Why this matters
A stuck task once ran for 1.5 days undetected, silently using up subscription quota.

Signs It's Working vs. Not

Working
  • You delegate without over-explaining
  • Output improves week over week
  • You spend less time on admin
  • Your AI anticipates your needs
Not Working
  • You re-explain context every session
  • You haven't given feedback in a week
  • You only use it when you "remember to"
  • Output quality is not improving

Things to Avoid

Every item below comes from a real incident.

1
If your AI offers to "restart the server" -- say no. It will knock the portal offline. Contact support instead.
Happened: AI restarted the portal to apply a change. Blank screen for 20+ minutes.
2
If your AI wants to "kill a process" -- contact support. This is IT plumbing. Let the support team handle it.
3
Don't wait hours if your AI stops responding. Contact support after 10 minutes.
Happened: "Rate limit" error persisted for hours. Actual usage was 7%. A restart fixed it instantly.
4
Don't run overnight tasks without a safety timer. Stuck tasks can run for days undetected.
5
Don't make changes right before something important. Customize during quiet moments, not before demos.
6
Don't share your portal login. Each person gets their own. Sharing means sharing everything -- files, emails, memories.
Simple rule
Your AI is great at work (writing, research, emails). For IT plumbing (restarting things, stopping things), contact support.

Common Mistakes

1
Not saving to permanent memory. The #1 reason subscribers feel PureBrain "doesn't remember." It does -- if you tell it to.
2
Treating it like a search engine. "What is X?" gets Wikipedia. "Prepare a brief on X for my Thursday meeting" gets something useful.
3
Skipping the handoff. Without it, you spend 15 minutes every session re-explaining context. That is 5+ hours a month wasted.
4
Not using all AI partners. If you have Partnered or Professional, using only one partner leaves capacity on the table.

Troubleshooting

What You See Why What to Do
Blank screen Portal stopped Force-refresh: Ctrl+Shift+R (Win) / Cmd+Shift+R (Mac). Still blank after 5 min? Email support.
"Rate limit" error Frozen session Wait 30 min. Still stuck? Email support.
AI forgot everything A compact happened Say: "Check your memory files and last handoff note."
Wrong colors/look Update reset styles Hard refresh. Then: "Fix my portal preferences."
Device hot/slow 3D animations "Turn off animations. Simpler backgrounds."
Emails not sending Password lost on restart "Check if your email passwords still work."
Good to know
The portal (the web page) and your AI (the brain) are two separate things. One can break while the other is fine. This helps when describing the problem to support.

Support

Email: witness-support@agentmail.to

Include: (1) what you see, (2) when it started, (3) what happened before. A screenshot is the fastest way to show the problem. Response: 15-60 minutes.

Tier Comparison

Awakening Partnered Professional Enterprise
AI Partners123Custom
AgentsCore setExpandedFull suiteCustom + dedicated
SupportDocumentationEmailPriority emailDedicated CSM
Strategy Sessions--MonthlyBi-weeklyWeekly
Custom Skills----AvailableIncluded
Best ForSolo professionals testing AI delegationConsultants and coaches scaling their practiceSmall firms with multiple workstreamsOrganizations needing tailored AI infrastructure

The Three Things That Matter Most

  1. Save to permanent memory. Every correction, every preference, every decision. If you don't save it, it didn't happen.
  2. Write handoffs. 5 minutes at session end saves 30 minutes at next session start. This is the compounding engine.
  3. Give feedback. Say what was good and why. Say what was bad and why. Then tell it to save the correction.

Quick Reference Card

Start of session"Check your memory and the last handoff before we start."
Save something"Remember that [detail]. Save to permanent memory."
After correction"Save this so you don't repeat the mistake."
Before compact"Save conversation, unfinished work, and write a handoff note."
End of session"Write a handoff: what we did, what's pending, decisions made."
After restart"Check that all your logins still work."
Supportwitness-support@agentmail.to