Build Your Specialized AI Team: Complete Setup Guide

Quick Answer: Create a project, define its purpose, generate role instructions with the Prompt Adjutant, upload reference documents, and let the Knowledge Graph compound your team’s expertise over time.

⏱ 15-20 minutes for initial setup

What This Guide Covers

You’ll learn how to transform Suprmind from a general-purpose AI tool into a highly specialized team of experts. By the end, you’ll have:

A dedicated project workspace with clear purpose
Five AIs that understand their specific roles
Reference documents as your team’s “training materials”
A Knowledge Graph that gets smarter with every conversation

The Setup Process

Create Your Project

Open Suprmind and click New Project in the sidebar.

Write a clear, specific description. This becomes the foundation for everything else.

Weak description:

Legal stuff

Strong description:

Commercial contract review for B2B SaaS agreements. Focus areas: liability clauses, indemnification terms, payment schedules, and termination conditions. Our company is the vendor. Contracts are typically 5-20 pages. We follow Delaware law unless specified otherwise.

The more specific your description, the better your AI team understands the job.

Generate Project Instructions

Now you’ll turn that description into proper instructions that every AI will follow.

Open the Prompt Adjutant (sidebar panel)
Paste something like this:

I need system instructions for a Suprmind project.

Project purpose: [paste your description from Step 1]

Create detailed instructions that:
– Define the core objective
– Specify what success looks like
– List what the AIs should always do
– List what they should avoid
– Define the output format preferred
– Include any domain-specific terminology

The Adjutant returns structured instructions
Copy the result

Example output:

PROJECT: Commercial Contract Review (B2B SaaS Vendor)

OBJECTIVE:
Review commercial contracts where our company serves as
software vendor. Identify risks, suggest improvements,
ensure compliance with standard terms.

ALWAYS:
– Flag unlimited liability exposure
– Check indemnification is mutual and capped
– Verify payment terms match our standard (Net 30)
– Note any auto-renewal clauses
– Highlight jurisdiction if not Delaware

NEVER:
– Approve contracts without flagging liability issues
– Skip fine print in exhibits/schedules
– Assume standard terms without verification

OUTPUT FORMAT:
1. Risk Summary (High/Medium/Low items)
2. Recommended Changes (specific redlines)
3. Questions for Legal Counsel
4. Overall Assessment (proceed/negotiate/reject)

Add Instructions to Your Project

Open your project
Click the Settings icon (gear)
Select Advanced Settings
Find Project Instructions
Paste your generated instructions
Save

Now every AI in every conversation within this project follows these rules.

Give Each AI a Specialized Role

This is where it gets powerful. Each AI can have its own personality and focus area within your project.

Go to Project Settings > AI Personalities tab.

For each AI, use the Prompt Adjutant to generate role-specific instructions:

Create a specialized role for [AI name] within a
commercial contract review project.

Project context: [brief project description]

This AI should focus on: [specific angle]

Generate instructions that define their expertise,
approach, and what unique perspective they bring.

Example AI roles for contract review:

AI	Specialized Role
Grok	First-pass scanner. Flag anything unusual. Quick pattern recognition. Check for recent regulatory changes that might apply.
Perplexity	Precedent researcher. Find relevant case law. Verify industry-standard terms. Cite sources for any legal claims.
Claude	Risk analyst. Deep-dive on liability, indemnification, IP assignment. Conservative interpretation. Flag ambiguities.
GPT	Structure checker. Ensure all required sections present. Verify internal consistency. Check cross-references.
Gemini	Synthesis and summary. Pull together all perspectives. Draft executive summary. Recommend next actions.

Paste each role’s instructions into the corresponding AI’s field in the AI Personalities tab.

Upload Your Reference Documents

Your AI team needs training materials. Go to Project Files and upload:

Standards and Guidelines:

Your company’s contract review checklist
Acceptable terms document
Red-line thresholds (what needs escalation)

Examples of Good Work:

3-5 contracts you’ve previously approved
Template agreements you prefer
Negotiation playbooks

Reference Materials:

Industry standard terms glossaries
Regulatory compliance summaries
Company policy documents

Supported formats: PDF, DOCX, TXT, MD, XLS

These become your project’s Vector File Database. The AIs can search and reference them automatically.

Start Working

Create a new thread. Attach the contract that needs review.

Ask your question:

Review this Master Services Agreement. Our company

(Acme Software Inc.) is the vendor. Flag risks,

suggest changes, and provide an overall assessment.

All five AIs respond in sequence. Each one:

Follows the Project Instructions
Plays their specialized role
Can reference your uploaded documents
Sees what the other AIs said before them

How Your Team Gets Smarter

Here’s what happens automatically as you work:

The Knowledge Graph Learns

A background process (called the Scribe) watches every conversation. It extracts:

Key entities: Company names, contract types, specific clauses you discuss
Relationships: Which terms connect to which risks
Decisions: What you approved, rejected, or flagged for escalation

This builds a graph of knowledge specific to your project.

Each Analysis Improves the Next

When you review your 10th contract, the AIs have context from the previous nine:

“Last time we saw this indemnification clause, you flagged it”
“This vendor had payment term issues in the August agreement”
“Auto-renewal was a deal-breaker in similar contracts”

They don’t just remember raw text. They remember patterns, decisions, and outcomes.

Self-Correction Built In

When one AI makes a mistake, others catch it:

Claude flags a liability risk
GPT notes the cap is actually in Exhibit B
Claude acknowledges and updates assessment

This happens naturally because each AI sees the full conversation history.

Real Example: Before and After

First Week

You upload a contract. The AIs give general analysis based on Project Instructions. Good, but generic.

First Month

After reviewing 15 contracts, the Knowledge Graph knows:

Your standard acceptable terms
Recurring issues with specific vendors
Which clauses always get negotiated
Your company’s risk tolerance

Third Month

The team anticipates your needs:

Flags patterns from past reviews automatically
Knows which issues escalated to legal counsel
References previous negotiations with the same counterparty
Suggests redlines based on what worked before

You’ve built institutional knowledge that compounds.

Optimizing Your Setup

When to Update Project Instructions

After you realize the AIs keep missing something
When your company policy changes
When you want to shift focus (e.g., more aggressive on payment terms)

Use the Prompt Adjutant each time. Tell it what needs to change.

When to Upload New Documents

New template agreements
Updated compliance requirements
Successful negotiation examples (so the team learns what “good” looks like)

Using @Mentions for Specific Tasks

Not every contract needs all five perspectives.

Quick standard agreement: @gpt @claude (structure check + risk scan)
Complex multi-party deal: All five AIs
Need precedent: @perplexity (cite case law and standards)

Non-mentioned AIs stay in context but don’t respond. Faster, cheaper, still smart.

Troubleshooting

AIs aren’t following instructions:

Check that Project Instructions are saved in Advanced Settings. They should appear at the top of every AI’s context.

Generic responses despite setup:

Upload more reference documents. The AIs need examples of “good” to calibrate against.

One AI keeps making the same mistake:

Update its specific role in AI Personalities. Be explicit about what it should avoid.

Knowledge Graph not helping:

It needs volume. After 10-15 substantial conversations, patterns emerge. Keep working.

Other Use Cases for This Approach

This same setup process works for:

Domain	Project Focus	Key Reference Docs
Medical Analysis	Reviewing research papers, treatment protocols	Clinical guidelines, approved studies
Investment Due Diligence	Evaluating opportunities, risk assessment	Investment criteria, past deal memos
Technical Architecture	Code review, system design	Style guides, approved patterns
Grant Writing	Proposal development, compliance	Successful proposals, funder guidelines
Content Strategy	Brand voice, editorial review	Style guide, approved examples

The pattern is the same: clear purpose, specialized roles, reference materials, and let the Knowledge Graph compound your expertise.

Summary: The 6-Step Setup

Create project with specific description
Generate Project Instructions using Prompt Adjutant
Paste instructions into Advanced Settings
Define AI roles in AI Personalities tab
Upload reference docs as training materials
Start working – the Knowledge Graph handles the rest

Your first analysis takes 15 minutes to set up. Your 50th analysis has a team that knows your preferences, your history, and your standards.

That’s how five AIs become your specialized expert panel.

Related Guides

Still need help? Use the feedback button in any conversation or contact support.