# Set up the two agents

## Choose one route
For the guided setup, use chatgpt Plus or Google AI Pro (often called Gemini Pro). Check the current plan and file limits before buying a new subscription. Some functions also exist on free plans; paid access gives higher limits rather than a guarantee of unlimited use. Your £49 purchase includes none of these subscriptions. Google Cloud is an alternative for a developer-built API setup, with separate billing; this pack does not require Cloud or API keys.

Bring a laptop, a text editor, one AI account, and a design or image tool you already use. A basic document editor is enough to prepare the files. Never enter API keys into these templates.

## What an MD file means
MD means Markdown. It is a normal text file with simple marks for headings and lists. `# My heading` is a heading; `- A point` is a list item. It is not an app and does not run code. The AI reads the words as instructions or reference material.

Unzip the pack. Open an .md file in a text editor such as Notepad, TextEdit in plain-text mode, or a Markdown editor. Make a working copy. Change the words inside square brackets, then save as UTF-8 text, keeping the .md extension. A file called brain.md.txt is a text file too, but rename it to brain.md for clarity. If a tool rejects .md, copy the contents into its context field or save an additional .txt copy. Uploading a ZIP does not guarantee the AI reads every file: extract it first.

## Prepare the Brain System
02_BUSINESS_BRAIN.md holds the offer, audience, facts, proof and next step. 03_VOICE_RULES.md holds how the business sounds. 04_VISUAL_RULES.md holds how it looks. Use only facts you can confirm. Write UNKNOWN for gaps; the agents must ask instead of inventing. Copy personal client details out of examples before uploading.

## Route A — chatgpt Projects
1. Open https://chatgpt.com/ and choose New project in the sidebar. Name it “Content AI Agent”.
2. Open the project menu and Project settings. Paste the text of 05_CONTENT_AGENT.md into the project instructions.
3. Add the three completed Brain files as project sources. Add two short writing examples you own, with one sentence explaining why each matches the voice.
4. Create another project named “Visual Directive Agent”. Paste 06_VISUAL_DIRECTIVE_AGENT.md into that project's instructions. Add the same Brain files and your own logo/reference images.
5. In each workspace, send: “Name your job, list the Brain files available, and name missing facts. Do not create anything yet.” Check the answer against the files.
6. Start the first real task inside the Content project. Use the handoff file to move approved writing to the Visual project yourself.

Project instructions and file sources are separate: uploading a role file alone is not the same as setting the project instructions. Separate projects do not automatically share updates or talk to each other.

## Route B — Gemini Gems
1. Open https://gemini.google.com/ on desktop. Open Gems and create a new Gem for the Content AI Agent. Labels can change; follow Google's current Gems help linked in 14_SOURCES_AND_TOOLS.md.
2. Put 05_CONTENT_AGENT.md into the Gem's instructions. Add the three completed Brain files as knowledge files, if that control is available in your plan.
3. Create the Visual Directive Agent Gem using 06_VISUAL_DIRECTIVE_AGENT.md, the same Brain files and your logo/reference images.
4. Test both with the file-check command above. Move approved writing between Gems using 08_HANDOFF.md.
5. If knowledge upload is unavailable or rejects .md, paste the completed Brain text at the beginning of a new chat with the role instructions. Keep that chat dedicated to the role. Repeat context after a new chat starts.

## First smoke test
Ask the Content Agent: “Using only the Brand Brain, write 80–120 words answering our chosen customer question. Name any missing fact before making a claim.”
Pass when it uses the correct reader, offer and CTA, and avoids invented proof. Then hand over the approved writing to the Visual Agent. Pass when it specifies size, layout, exact text, colours, logo placement and a final review check.

## Keep both agents current
Change the master Brain files first. Add a version/date. Replace the old sources in both workspaces. Ask each agent which version it is using. Old chats can retain old context: begin a fresh task after major changes. Save approved outputs outside the AI workspace as well.

## Basic privacy
Keep API keys, passwords, private customer records and card data out of prompts. Use fictional or anonymised examples. Check each provider's data controls and your client agreement before uploading client work. Review output before sharing.