Welcome to the colony.
Here's how to run it well.

1 Your first 5 minutes

In this section: create your account · sign the beta agreement · connect one AI provider · meet Stinger

Four things happen once, the first time you ever open CortexHive. In order:

1. Create your account. On the sign-in screen, tap "No account yet? Create one →". Enter your invite code, your email, and a password (10+ characters), then tap Create Account.
2. Read and sign the beta agreement. One screen, five short clauses: confidentiality, beta status, "you're talking to AI, verify anything important," what data is stored, and governing law. Scroll to the end, check the box, then type your full name: that's your signature, recorded with your email, IP address, and a timestamp. It appears once per account.
3. Connect one AI provider. Claude, ChatGPT / OpenAI, or Gemini, whichever you already use (DeepSeek, Moonshot / Kimi, and xAI can be added any time in Settings). Sign in with Claude and Sign in with ChatGPT are one click; OpenAI and Gemini also connect with an API key. You can't skip this step: CortexHive runs on your account, not ours, so nothing works until one is connected. Don't want to use a paid account yet? Skip ahead to Costs & control for the free NVIDIA path, no credit card required.
4. Meet Stinger. Your chief of staff opens the conversation, asks a few questions about you and your goals, and gets your starter team (25 specialists plus Stinger) moving.

The Connect your AI gate: Claude's "Sign in with Claude" needs no key; OpenAI and Gemini connect with a pasted API key. NVIDIA's free tier is right below, no credit card.

2 Set up for best results

In this section: tell Stinger about your business · bulk-import documents · connect 2-3 Power-ups

The single biggest lever for good output is context. Give the hive real information before you ask it to work.

1. Tell Stinger about yourself. Right after connecting, answer Stinger's setup questions (by chat or a quick form) about your business, your goals, and how you like things done. This shapes your starter team and your first tasks, and feeds the Activation Journey in the next section.
2. Import your documents. Open Documents in the left sidebar and drag files straight onto the page: up to 20 at once, .txt, .md, or .pdf. The hive reads what you give it before it answers, instead of guessing. See a real, populated Documents view in Section 5, The Second Brain.
3. Connect 2-3 Power-ups you actually use. Mission Control → Power-ups can wire 500+ tools (Gmail, Google Calendar, Slack, Notion, and more): 34 live today, and more every day. Two or three real ones beat ten you'll never touch.

3 The Activation Journey

Flagship · not a chatbot, a workforce with a mission

Right after your first quick start, Stinger nudges you toward one thing: a real conversation that turns into a real plan. This is what makes CortexHive different from a chat window: it ends in agents doing specific work, owned by name.

Part 1: the deep interview

Stinger asks about you across seven chapters: personal goals, work, health, coaching, things to learn, preferences, and short and long-term goals. Open it from the Deep interview item in the sidebar, or the "Deep interview" quick action, any time.

1. Answer one chapter at a time. Type or use the mic; each chapter saves the moment you move on.
2. Stop and come back whenever. The interview is resumable: reopen it later and it lands you back on the first unanswered chapter, not the start.
3. Review or edit any answer. "Review all answers" shows everything you've told Stinger in one place, each one editable.

Deep interview: seven chapters, one conversation at a time, saved and resumable.

Part 2: the activation plan

Once the interview is done, Stinger runs your answers, your north star, and the real roster of agents in your account through the flagship model to write your Activation Plan: a chief-of-staff briefing, not a chat reply.

A real activation plan: first moves owned by name, not a list of generic suggestions.

Part 3: your north star

Mission Control's briefing carries one more piece: a banner where you set the single goal the whole hive works toward. It shows once, as a prompt, until you set it, then it steps aside for good, you can always change it again from the briefing. You can see it in its opening prompt state at the top of the Daily driving screenshot next.

4 Daily driving

In this section: read your briefing · talk to the voice chief of staff · durable threads in Agent Chat · team coordination mode

Two habits cover most days: check Mission Control, then talk or type. Go deeper in Agent Chat when a conversation deserves its own thread.

1. Open Mission Control. Dashboard gives you a briefing: what the hive did, what needs your attention, over the honeycomb background that marks every Mission Control screen. It's cached for 3 minutes, so glancing at it twice in a row doesn't burn a fresh AI call.
2. Talk or type to Stinger. The composer takes plain text, or tap the mic. OpenAI accounts get live duplex voice (talk and get answered in real time); every provider gets standard voice (say something, hear a spoken reply). Typing always works.
3. Let Stinger act, hands free. This isn't voice-to-text into a chat box: Stinger can search the second brain, write documents, and create or assign Cortex OS tasks to any agent by name, by voice. It confirms out loud what it did in one short sentence, and asks first before anything irreversible.
4. Open the full Agent Chat app for durable threads. Click Agent Chat in the sidebar (or go straight to https://app.cortexhive.ai/chat): pick any agent on the left, start a named conversation, and it's saved so you can pick the thread back up later, on any device. Each conversation has its own model picker in the header, so a quick question and a heavyweight plan can run on different tiers.
5. Choose how a conversation coordinates the team. Open the three-dot menu next to rename and delete in a conversation's header and pick Live in chat to watch agents hand work to each other as it happens, or In the background to keep the thread quiet and collapse the coordination into a "Team activity" row you can expand any time. Either way the work happens the same; this only changes what you watch.

Mission Control's Dashboard: the north star banner up top, Stinger's orb and greeting below, the composer at the bottom.

The voice chief of staff: the same orb listens, thinks, and speaks, and now executes and delegates instead of only replying.

Agent Chat: a durable, one-on-one thread per agent, saved so you can pick it back up on any device.

The team coordination menu, next to rename and delete: Live in chat vs. In the background, with the per-conversation model picker beside it.

5 The Second Brain

In this section: what the second brain is · how to ask for it · wikilinks and backlinks

Every agent in the hive can reach into one shared memory: your documents, your Cortex OS tasks, your deep interview answers, and your goals, ranked and connected, in English or Spanish.

This isn't a separate screen, it's a capability every agent carries (the tool is called search_workspace). Ask any agent a question like "what do we know about the wholesale pricing model" or "what did we decide about the Spring Blend launch", and the answer comes back grounded in your real workspace, with the connected notes it pulled from named, not invented.

1. Write documents that reference each other. Link one document to another with double brackets, [[Like This]], Obsidian-style. No special syntax to learn beyond the double brackets.
2. Ask, don't search. Instead of hunting through folders, ask an agent directly: it ranks across documents, tasks, interview knowledge, and goals, and follows the links between them.
3. Trust the connections are real. A document that links three other documents shows up as a small, real web, the same shape as an Obsidian graph, built from what you actually wrote.

Documents, the second brain's raw material: each card's preview shows real [[wikilinks]] to the documents it connects to.

6 The Jury

In this section: what The Jury is · the four modes · reading the cost preview before you convene

Before you ship a plan, a post, or a pitch, get an honest read from a simulated audience instead of guessing.

The Jury seats dozens of distinct customer personas (skeptics, bargain hunters, early adopters, enterprise buyers) who react to your idea independently, with no groupthink. A foreman reads every reaction and delivers one decisive verdict. Connect two or more AI providers and the panel spreads round-robin across them, GPT, Claude, and Gemini jurors reacting to the same input, so the result is not just one model's opinion of itself.

1. Pick a mode. Vote, Debate, Brainstorm, or Test Hypothesis: each changes the prompt and the shape of the answer you get back.
2. Paste what you want judged. A plan, a post, a pitch, an offer, or a hypothesis. Optionally name your audience, or specific personas to seat first.
3. Choose panel size and juror model. Panel size is Quick (about 8 jurors), Standard (about 15), or Full (about 30). Juror model is Fast (recommended), Balanced, or Flagship.
4. Read the cost preview before you convene. The "Before you convene" panel shows the real call math, roughly one model call per juror plus one foreman call, on your connected account.

The Jury: four modes, a plain-language explainer, and a cost preview before anything runs on your account.

7 The Hive

In this section: browse the roster · when to call The Challenger · hire a new specialist · make an agent yours

You start with 25 specialists plus Stinger as chief of staff: 26 agents, day one.

1. Browse the roster. Mission Control → The Hive shows every agent and what it's scoped to do, in Cards, List, Bench, Alumni, or a full Architecture graph view.
2. Ask directly, or let Stinger route. Talk to a specialist by name, or describe the task to Stinger and let it hand off.
3. Use The Challenger before a real decision. Give it a plan, a draft, or a number, and it argues against it hard, then tells you plainly if the plan actually holds up.
4. Need a role that doesn't exist yet? Tell Stinger. It hands the request to The Architect, which designs and builds the new specialist from a plain-language description, no code, no special screen.

The Hive: every agent, grouped by division. Click a card to chat, inspect, or reassign it.

Make an agent yours

Every stock agent is built in two layers: a core (its role, skills, and operating rules) that is product-managed and keeps improving on its own, and a customization layer that is entirely yours. The core never changes because you wrote something; your customization rides on top of it and is never overwritten.

1. Open an agent's profile. Click Inspect on any agent card to open its drawer, on the Profile tab by default.
2. Tell it about your world. In the Customization box, describe your business, your brand voice, and how you like things done, plain language, no special syntax.
3. Save it. Click Save customization. From then on this agent applies your notes on top of its core role, every time.

The Customization editor: your notes about your business layer on top of the agent's managed core, and nothing you write is ever lost on an update.

8 Cortex OS

In this section: Tasks and Projects · Clients and Contacts · the Follow-Up Inbox · Opportunities

Cortex OS is where the hive's real work lives: tasks, projects, deals, and the people behind them, all in one operating system your agents read and write directly.

1. Open Cortex OS. Click Cortex OS in the Mission Control sidebar, or go straight to https://app.cortexhive.ai/crm.
2. Work the Tasks board. Group by project, filter by status or priority, and every task an agent creates or completes shows up here in real time, the same list Stinger reads when you ask "what's on my plate today."
3. Track deals in Opportunities. A Kanban pipeline by stage, Lead In through Discovery, Proposal, Negotiation, to Closed Won or Closed Lost, each card carrying its value.
4. Let agents update it for you. Any agent can create, read, and update Cortex OS tasks as a tool, and can be delegated a task by name straight from voice or chat, see Daily driving.

Cortex OS Tasks: grouped by project, every status from Not Started to Done, assignee and priority on every row.

Cortex OS Opportunities: the deal pipeline, stage by stage, with real dollar values on every card.

9 Getting real work out

In this section: run a Recipe · save it as a workflow · journal with Scribe · automate the repeat ones · build a custom dashboard

Recipes, workflows, and automations are the same idea at three levels of commitment.

1. Start from a Recipe. The Recipes tab has ready-made templates grouped by category, planning, outreach, finance, marketing, and more. Run one once to see what it produces.
2. Save what works. Turn a recipe, or anything you build yourself, into a named, repeatable workflow with Save.
3. Automate the ones you'd run every week anyway. A saved workflow becomes an automation on a schedule or a trigger (needs a connected Power-up).
4. Keep a running journal. Journal is a stream-of-consciousness dump to Scribe: type or talk, in English or Spanish, and Scribe files it, tags it, and flags anything that should become a task.
5. Find the output. Anything written lands in Documents. Reports gives you a live rollup of hive activity (tasks, workflow runs, automations), no writing required.

Recipes: run once to see the output, Save to keep it as a workflow, Automate to put it on a schedule.

Journal: dump it here, Scribe sorts it, files it, and flags anything that should become a task.

Build a custom dashboard

Reports → Dashboards lets you build your own live boards, Shopify-style, from your account's real data: a number, a trend line, a leaderboard, a table, a heatmap, or a status split. Arrange them your way, then turn them into a report on a schedule.

1. Create a dashboard. Reports → Dashboards → + New dashboard, give it a name.
2. Add cards. + Add card, pick a type and a real data source (spend, tasks done, calls per day, top agents, open tasks, and more), set its width, and it appears on the board immediately with live numbers.
3. Arrange it your way. Drag a card's grip handle to reorder, or use its menu to resize.
4. Turn on automated reports. Set the Reports dropdown to Daily or Weekly, or click Generate report now for an on-demand one, saved straight to Documents.

Reports → Dashboards: your own boards, built from live account data, with scheduled reports on top.

10 Costs & control

In this section: BYOK billing · the free NVIDIA path · route models on purpose · read real numbers in Analytics

You're always spending your own AI account, never ours. Here's how to see it, steer it, or try CortexHive for nothing first.

1. BYOK, bring your own key. CortexHive runs on the Claude, OpenAI, Gemini, or other account you connect. What you spend goes straight to that provider, at their price, with no markup.
2. No card yet? Start on NVIDIA's free tier. Settings → AI Connections lists NVIDIA alongside Claude, OpenAI, and Gemini: every developer account starts with an allowance of free API credits, no credit card needed, enough to try the whole product. NVIDIA's own terms scope this tier to evaluation, not running your real business on it, so move to a paid provider once you're past trying it out.

   Watch: generating your free NVIDIA key (about 2 minutes)

3. More providers, if you use them. DeepSeek, Moonshot/Kimi, and xAI (Grok) connect the same way, each with a step-by-step "how to connect" panel, and a short setup video for the providers that have one.
4. Route models on purpose. Pick a model per agent, or per conversation (see the model picker in Agent Chat), when it matters: heavier reasoning for a real plan, a lighter model for volume work.
5. Check Analytics. The Analytics tab breaks down calls, tokens, and cost, with a 30-day activity chart and an ROI multiple: real numbers, not estimates.

Video is a separate, optional connection. Settings → Media generation takes a Higgsfield API key and turns on video: your agents can animate a picture into a short clip and it lands in your workspace as a download. It is not an AI account and does not power chat, so it does not replace one of the providers above. Generation runs on your own Higgsfield plan and spends your own Higgsfield credits, CortexHive does not resell them, and every render asks you to approve it first. Content you generate through a connected Higgsfield account is handled under Higgsfield's own terms, including their use of inputs and outputs to train their models. CortexHive never sells your data and never trains on it.

Settings → AI Connections: every provider CortexHive supports, including NVIDIA's free, no-card tier.

Analytics: cost, calls, tokens, and ROI, refreshed from your account's real usage.

11 Troubleshooting, quick hits

In this section: the most common hiccups · Shortcuts & Capabilities · where the full FAQ lives

Most hiccups are self-explanatory once you know what they mean.

• Voice says "using standard voice." Live duplex hit a hiccup, often a VPN blocking the audio channel, and fell back automatically. Reload to retry immediately, or toggle your VPN off.
• Who gets which voice? Only OpenAI supports live duplex today. Everyone gets standard (turn-based) voice: OpenAI accounts hear GPT voices, Gemini accounts hear Gemini's native voice, Claude and everyone else hear ElevenLabs premium voices.
• A provider key stops working. Reconnect it from Settings → AI Connections. Nothing else about your workspace changes.
• My activation plan feels generic. Finish all seven chapters of the deep interview first and set a north star; the plan is only as grounded as what it's given.
• Forgot your password. Click Forgot password? on the sign-in screen. We email a reset link that is valid for one hour and can only be used once.

Look up anything: Shortcuts & Capabilities

Shortcuts & Capabilities in the sidebar is a live reference, not a tour: every keyboard shortcut, plus exactly what each agent can do and how to trigger it, from talking to Stinger to connecting a Power-up. Skim it once and you'll stop wondering whether CortexHive can do something.

Quick picks: ⌘K / Ctrl+K command bar, Esc close or cancel, Enter to send and Shift+Enter for a new line, ⌘S / Ctrl+S to save a document or your Hive layout.