DOCUMENTATION
Cortex AI, from A to Z
No coding needed. This guide walks you through everything — getting your own API keys, creating your first AI agent, teaching it about your business, connecting it to Gmail/Slack/Notion, and putting it live on your website. Follow it top to bottom if you're brand new.
1. What is Cortex AI?
Cortex AI lets you build your own AI employee — an "agent" — without writing a single line of code. You give it a job (answer customer questions, qualify leads, summarize documents), teach it about your business by uploading your own files, optionally let it use apps like Gmail, Slack, or Notion, and then put it live as a chat link, a widget on your website, or an API other software can call.
📸 Screenshot: the main dashboard showing a list of created agents.
2. Creating your account
Go to the homepage and click Start free trial.
Enter your name, email, and a password. This automatically creates your own private workspace (called an "Organization") — nobody else can see your agents or documents.
You'll land straight on your dashboard, ready to build your first agent.
📸 Screenshot: the signup page (name, email, password fields).
3. Getting your API keys
An AI agent needs a "brain" to think with — that brain is a language model from a provider like OpenAI, Anthropic, Google, or Groq. To use most of these, you need your own API key from that provider (think of it like a password that lets Cortex AI talk to that company's AI on your behalf, and bills you directly for what you use). Here's where to get each one:
🟢 Groq — free tier available
You can skip this entirely and use Cortex AI's free shared key (limited messages per day). Want your own unlimited key instead?
Go to console.groq.com → sign up → API Keys → Create API Key → copy it.
🔵 OpenAI
Requires your own key — add billing on OpenAI's side to use it.
Go to platform.openai.com → log in → API keys (left sidebar) → Create new secret key → copy it immediately (it's shown once).
🟣 Anthropic (Claude)
Requires your own key and billing set up.
Go to console.anthropic.com → log in → API Keys → Create Key → copy it.
🟡 Google Gemini
Requires your own key.
Go to aistudio.google.com → log in → Get API key → Create API key → copy it.
📸 Screenshot: the "API key" field on the Create Agent form, showing where the key gets pasted.
4. Free tier vs. bringing your own key
When you create an agent, you pick a model provider. Every provider needs your own API key, except one:
- ✅ Groq — leave the API key field blank and your agent runs on Cortex AI's shared free key, capped at a limited number of messages per day. Perfect for trying things out.
- 🔑 OpenAI / Anthropic / Google Gemini — you must paste your own key. There's no shared fallback for these, since they're paid providers.
You can always start on the Groq free tier, and switch to a paid provider (with your own key) later from the agent's edit screen, once you're ready for a more powerful model or higher usage.
5. Creating your first agent
There are two ways to create an agent:
Option A — Quick create (fastest)
On your dashboard home screen, type what you want the agent to do in the one-line box — e.g. "A customer support bot that answers from my FAQ and refund policy docs".
(Optional) Click Advanced options to set the name, pick a model provider, and paste your API key if you're not using Groq's free tier.
Attach any documents right there if you already have them, then hit send. Your agent is created instantly.
📸 Screenshot: the "describe your agent" box on the dashboard home screen, with Advanced options expanded.
Option B — Full form (more control)
Go to Agents → New agent.
Fill in the name, pick an agent type (Customer Support, Sales, Research, etc.), and write plain-language instructions in System instructions — this tells the agent who it is and how to behave.
Choose a model provider and model, and paste your API key (or leave it blank for Groq's free tier).
Click Create agent.
📸 Screenshot: the full "Create an agent" form (name, type, system instructions, provider, model, API key).
6. Teaching it about your business (Knowledge Base / RAG)
"RAG" just means your agent reads your own documents before answering, instead of guessing. This is what makes it answer from your FAQ, policies, or product docs — not generic internet knowledge.
Open your agent, and find the Knowledge base section.
Upload files — PDF, Word (DOCX), plain text, CSV, or Markdown all work.
Wait a moment while it's processed (you'll see a status like "Processing" turn into "Ready"). Once ready, your agent can pull facts from it and will mention which document an answer came from.
📸 Screenshot: the knowledge base tab with a document uploaded and its status ("Ready").
7. Testing in the Playground
Before showing your agent to real customers, test it yourself. Every agent has a built-in Playground — a private chat window only you can see — where you can ask it questions exactly like a visitor would, and see how it responds, in real time.
📸 Screenshot: the Playground chat window with a sample question and answer.
8. Connecting apps (Connectors)
Connectors let your agent actually do things in other apps, not just talk. Go to Connectors in the sidebar to set these up, then link the ones you want to a specific agent from that agent's settings.
📸 Screenshot: the Connectors page showing all available integrations (Gmail, Slack, Notion, etc.).
Available connectors
- 📧 Gmail — read and send email from a connected mailbox.
- 📁 Google Drive — search and read your Drive files.
- 📝 Notion — read and update Notion pages.
- 💬 Slack — read channels and post messages via a bot.
- 📊 Google Sheets — read and write rows in a spreadsheet.
- 📅 Google Calendar — check availability and create events.
- 🧡 HubSpot — look up and update CRM contacts/deals.
- 📱 WhatsApp Business — send and receive WhatsApp messages.
- 🎨 OpenAI (image generation) — let the agent generate images.
- ▲ Vercel — let the agent build and deploy a live website (one or more pages) and hand you back the URL. You'll need a Vercel account (free) and a Personal Access Token — click "Create Token", give it any name, and paste the value shown into the Vercel card on the Connectors page.
- 📥 Incoming Webhook — let outside systems (a website form, Zapier, Make, n8n) trigger the agent. See the next section.
- 🌐 HTTP Request — let the agent call any external web address you specify.
Click Connect under the app you want, and follow the on-screen steps (usually logging in to that app and approving access).
Open your agent's settings and link the connector to it, choosing whether it can only read or also send/write.
📸 Screenshot: an agent's connector settings, showing a toggle for read vs. send access.
9. Incoming Webhooks (letting outside systems trigger your agent)
Most connectors let your agent reach out to other apps. An Incoming Webhook works the other way around — it gives you a private web address that any outside system can send information to, which instantly wakes up your agent to handle it and reply.
On the Connectors page, find Incoming Webhook and click Generate webhook (optionally set a shared secret for extra security).
Copy the URL shown — it's only shown once.
Link this connector to the agent you want it to trigger.
Paste that URL into whatever should trigger your agent — your website's contact form, a Zapier/Make/n8n scenario, or any other tool that can send data to a web address.
📸 Screenshot: the one-time "copy this URL" banner after generating an Incoming Webhook.
10. Going live
Once you're happy with how your agent answers, open its Deployment tab. You'll get three ways to put it to work, all at once — no separate "publish" step:
- 🔗 Shareable link — a public chat page you can send to anyone, or keep private behind a secret access token.
- 🧩 Embed widget — a small chat bubble you can add to your own website, restricted to only the domains you allow.
- 🔌 API endpoint — for developers who want to call your agent from their own software.
📸 Screenshot: the Deployment tab showing the public link, embed code snippet, and a welcome-message/branding preview.
11. Viewing conversations
Every message your agent sends or receives — whether from the Playground, the public link, the embed widget, or a webhook — is saved. Open an agent's Conversations tab to read through past chats and see how it's actually performing with real questions.
📸 Screenshot: a list of past conversations for one agent.
12. Plans & billing
Cortex AI's own subscription covers hosting your agents, the dashboard, and the free Groq tier — it does not include the cost of your own OpenAI/Anthropic/Google API usage, which those providers bill you for separately based on how much you use them. See the Pricing page for current plans.
13. Troubleshooting & FAQ
My agent says it needs an API key.
You picked a provider other than Groq without pasting a key. Open the agent's edit page and either add your key for that provider, or switch the provider to Groq to use the free tier.
My agent says the free tier limit is reached.
Groq's free tier is capped at a limited number of messages per day. It resets at midnight, or you can add your own Groq API key on the agent to remove the cap entirely.
My agent gives wrong or outdated answers.
Check the Knowledge Base — it can only answer from what you've actually uploaded. Update or re-upload the relevant document.
A connector (like Slack) isn't working.
Make sure the connector is both connected and linked to that specific agent, with the right read/send permission, and that any bot (e.g. Slack) has actually been invited to the channel you're asking about.