Get started with Copilot

Copilot is the AI chat built into your Hivenue workspace. You open it from the sidebar and talk to Claude, Anthropic's AI, the way you would on claude.ai. The difference is that here Claude can reach your workspace's tools: it reads your Meta Ads performance, checks your automation rules, scores creatives, and, if you allow it, acts on campaigns for you.

Copilot is bring-your-own-key: you connect your own Anthropic API key and pay Anthropic directly. Your prompts go from your workspace straight to Anthropic, with no intermediary storing them, and you see exactly what you spend.

Before you start

• An active Hivenue workspace. You need the admin or owner role to connect the key; members can chat once it is set up.
• Your own Anthropic API key (created below).
• A Meta connection is only needed if you want Claude to work on your campaign data. Plain chat works without it.

Step 1: Create an Anthropic API key

1. Go to console.anthropic.com and sign up or log in.
2. Add a payment method. Anthropic works on prepaid credit, so load an initial amount.
3. Open Settings, API Keys and click Create Key. Name it something you will recognize later, like the name of your workspace.
4. Keep the default scope. Copilot only needs the standard messages endpoint, no admin or billing permissions.
5. Copy the key immediately. It is shown once, starts with sk-ant-, and if you close the dialog without copying you simply create a new one and revoke the old.

Treat the key like a bank password: never paste it into chat tools, email or shared docs. Whoever holds it can spend on your account until you revoke it.

Step 2: Connect the key to Hivenue

1. In your workspace, open Settings, Integrations, AI. Only admins and owners see the key form; members see usage statistics only.
2. Paste the key into the Anthropic API key field and click Save. Hivenue validates it against Anthropic with a free call that consumes no tokens; there is also a Test connection button if you want to check first.
3. Three outcomes are possible:
   • Valid: the key is encrypted with AES-256-GCM and stored. A green badge confirms it, and the Copilot page is live.
   • Invalid: the key is wrong or revoked. Check it on the Anthropic console.
   • Rate limited: Anthropic is temporarily refusing calls. The key is saved anyway; retry the validation in a few minutes.

After saving, the key is never shown in full again, only a masked version. To replace it, paste a new one; to remove it, click Delete. The key is decrypted only on the server for the duration of a chat turn, is never sent to the browser and never written to logs. See the privacy page for the whole policy.

Step 3: Set your guardrails

Under the key form, four admin controls keep spending predictable:

Control	What it does	Default
Default model	Which Claude model Copilot uses. Sonnet is the balanced default; Haiku costs roughly a tenth, Opus is the top model at a higher price	Sonnet
Daily message cap	Hard ceiling on messages per 24 hours for the workspace	500
Max tool iterations	How many tool calls Claude may chain inside one reply	10
Hourly rate limit	Maximum messages in 60 minutes	100

Changes apply from the next message onward. The caps exist so an accidental loop or an enthusiastic teammate cannot burn through your Anthropic credit in an afternoon.

Step 4: Start chatting

Open Copilot from the sidebar. Three suggestion cards give you ready-made starting points: analyze your best and worst creatives of the last 30 days, find low-ROAS ads and propose actions, or set up your brand voice. Click one, edit the prompt if you like, and send.

Claude's toolbox splits in two:

• Read tools, always available: campaign insights, automation rules and their runs, past AI generations, brand voice, remaining AI budget, and the audit history.
• Write tools, off by default: pause or resume a campaign, adjust budgets, send a notification, score a creative, generate headlines or image variations, update the brand voice.

Write tools stay disabled until an admin flips the AI write access toggle on the AI settings page. Even with the toggle on, only admins and owners can use them; if a member asks Claude to pause a campaign, Claude answers that it lacks the authorization. Every tool execution appears in the chat as a collapsible card with the tool name, input and output, so nothing happens silently.

A practical habit: leave write access off except when you are actively optimizing with the chat open. All the analysis works either way.

Good to know

• Billing: Anthropic charges your prepaid balance per token. Your Anthropic dashboard is the ground truth; the Hivenue AI settings page shows today's messages, tokens and cost plus a 7-day chart. Prompt caching is on by default, which noticeably cuts the input cost of active sessions.
• Refusals: Claude can decline requests that look like sensitive claims, prompt injection or targeting of protected categories. Rephrase descriptively, drop words like "guaranteed", and retry. A refusal costs only a handful of tokens.
• The chat is transient by design. Close or refresh the page and the conversation is gone. Hivenue stores only metadata (token counts, model, status), never your prompt or Claude's replies. There is no sharing, no export, no image upload and no voice in this version, and one API key serves the whole workspace.
• Model choice is per message. Switch the default model in settings and the next message uses it; earlier messages keep their own record.
• If your key leaks, revoke it on the Anthropic console (instant), create a new one, and paste it into Hivenue. The settings page flags the dead key so you notice fast.