Quick start

This guide gets you from zero to a live, AI-drafted reply in roughly 15 minutes. You will create a workspace, connect a channel, paste in your AI key, and send a real conversation through it.

Before you start

You will need:

• An admin email address you can receive mail at.
• Either an Anthropic or OpenAI API key. Ochre is BYOK, so you bring the key and pay the provider directly. Both providers offer free trial credit if you do not have an account yet.
• 15 minutes.

Step 1: Create your workspace

Go to ochrehq.com and sign up with your work email. Pick a workspace name and a subdomain. The subdomain becomes your help center URL by default (you can swap it for a custom domain later — see Help center domain).

Your account becomes the Owner. Owners can do everything, including billing and workspace deletion. You can transfer ownership later if needed.

Step 2: Add your AI key

Open Settings, then Integrations, then AI. Paste either an Anthropic or OpenAI key. Ochre validates the key in place and shows you the model list it can call.

Pick a default model. Sonnet-class models are the sweet spot for most teams: smart enough for hard tickets, fast enough for live chat. You can change it later or set per-channel models. See Choosing a model for the breakdown.

Set a monthly spend cap. The AI stops drafting when you hit the cap and Ochre alerts you 24 hours before. See Spend caps.

Step 3: Connect your first channel

Most teams start with email. From Settings, then Channels, choose Email channel. Two options:

• Forward an existing address — set up a forwarding rule in Gmail, Outlook, or your provider that points support@yourcompany.com at the Ochre address shown on screen. Replies will go out from a generic Ochre address until you add your domain.
• Custom domain — add your sending domain and follow the DKIM, SPF, and MX records we generate. Replies go out as support@yourcompany.com and inbound mail lands directly in your inbox. Full walkthrough: Custom domain email.

If you want a chat widget instead, see Chat widget. For Slack Connect channels, see Slack Connect.

Step 4: Seed the AI brain

The AI agent draws on your knowledge base to answer customers. You have three paths:

1. Connect Notion or GitBook — point the AI at an existing docs space and it ingests automatically. See Connecting Notion.
2. Write articles in Ochre — open Knowledge base in the sidebar and start authoring. The first article takes 5 minutes. See Authoring articles.
3. Skip for now — the AI will draft from conversation context alone. It still works, but answers are weaker until it has reference material.

Step 5: Invite one teammate

Inbox tools are more fun with two people. Open Settings, then Team, and send an invite. Pick a role:

• Admin — full access except billing and deletion.
• Agent — handles conversations.
• Light agent — read-only, can leave internal notes. Free, not billable.

See Roles explained for the full matrix.

Step 6: Send a test conversation

Email your support address from a personal account, or pop the chat widget on a staging page and send a message. Within a few seconds you will see:

• A new conversation in Unassigned.
• A label for topic, priority, and confidence.
• An AI-drafted reply in the composer.

Read the draft. If it is right, hit send. If not, edit it and send. Either way, the AI learns from your edits and gets closer next time. See Drafting modes and Learned QA.

Step 7: Decide how aggressive the AI should be

Three modes:

• Draft-only — you review every send. Safest, slowest.
• Copilot — one-click send on the AI draft. Standard for most teams.
• Autopilot — AI sends on its own when confidence is above your threshold. Fastest, requires guardrails.

Start with copilot for a week, then turn on autopilot for high-confidence how-to questions once you trust the voice. Full setup: Autopilot.

What to do next

You are live. From here:

• First conversation — a deeper walkthrough of the inbox.
• Inviting your team — bring the rest of support in.
• Inbox filters — learn the keyboard-driven workflow.

If something is not landing, jump to Troubleshooting overview or contact support.