BYOK: bring your own Anthropic or OpenAI key

Ochre is bring-your-own-key. You connect an Anthropic key, an OpenAI key, or both. The AI runs on your provider account. You pay the provider directly. Ochre never marks up tokens.

This is on purpose. We charge a flat seat price. Token costs are between you and Anthropic or OpenAI. The numbers in your provider dashboard match the numbers we show in your AI receipts.

Why BYOK

Three reasons:

1. Honest pricing. No vendor in the loop quietly adding margin to every reply. The math typically lands at $0.005 to $0.01 per resolution on Haiku 4.5 or GPT-4o-mini, versus the $1.50 to $2.00 per resolution incumbents charge.
2. Your data, your contract. The terms you signed with your model provider apply. If you have a zero-retention agreement with Anthropic or OpenAI, that agreement is in force.
3. Volume discounts pass through. Enterprise rates from your provider show up directly in your bill.

Supported providers

Anthropic and OpenAI only. Gemini, Mistral, Cohere, self-hosted, and OpenRouter-style aggregators are not supported today.

What you need

• An Anthropic API key with access to the Claude models you want.
• An OpenAI API key with access to the GPT or o-series models you want.
• You can connect either, both, or one now and the other later.

You do not need a separate key per agent. The workspace uses a shared key.

How to connect a key

1. Open AI → Keys.
2. Click Add key → on the Anthropic or OpenAI card.
3. Paste the key (sk-ant-… for Anthropic, sk-proj-… for OpenAI).
4. Pick the default model.
5. Click Verify & save. Ochre validates the key against the provider before storing it.

Keys are encrypted at rest. Only Owners and Admins can add, view, or rotate them. Agents see "Connected" or "not configured", never the secret.

What models are available

The picker covers the current shipping catalog from each provider:

Anthropic. Claude Opus 4.7, Claude Opus 4.5, Claude Sonnet 4.6, Claude Sonnet 4.5, Claude Haiku 4.5, plus the 3.x legacy models for compatibility.

OpenAI. GPT-5, GPT-5 mini, GPT-5 nano, GPT-4.1, GPT-4.1 mini, GPT-4.1 nano, GPT-4o, GPT-4o mini, plus the reasoning models o3, o3-mini, o4-mini, o1, o1-mini.

Pick a default per provider on the same modal. You can override per channel later. See Choosing a model.

What it costs

Per resolution typically lands between $0.005 and $0.01 on the cheaper, faster models. A "resolution" is the full work of one AI reply: classification, retrieval, drafting, and labeling. Cost depends on:

• Model. Haiku and the GPT mini/nano tier are cheapest, Opus and GPT-5 are the most expensive.
• Conversation length. Longer threads mean more input tokens.
• KB retrieval. Larger context windows pull in more articles.

Every reply has a receipt with the exact tokens and dollars. Set a monthly cap in Spend caps and alerts.

Two keys, automatic fallback

Connect both providers and Ochre uses one as primary, the other as backup. If your primary returns a rate limit or 5xx, the next reply switches providers without dropping the ticket. Read Multiple keys + automatic fallback.

Rotating a key

Click rotate on the key card under AI → Keys. Paste the new key, verify, save. Rotation is hot: in-flight requests finish on the old key, new requests use the new one. You do not need to pause AI to rotate.

If you click Disconnect with no fallback configured, AI features pause until you connect a new key. Your inbox keeps working as a normal helpdesk.

Permissions and visibility

• Owners and Admins can add, view, rotate, and disconnect keys.
• Agents and Light agents see only whether AI is on, not the keys themselves.
• The org-level audit log records every key add, rotation, and disconnect event.

Common questions

Does Ochre proxy through its own account? No. Calls go from Ochre's backend to your provider on your key.

Will my provider see customer content? Yes, the same as any AI helpdesk. Use your provider's privacy controls (zero retention, data processing agreements) to set the terms.

What if my provider account is suspended? Your inbox keeps working without AI. Connect a new key or fall back to the other provider, and drafting resumes on the next message.

Can I use a self-hosted or third-party model? Not today. Anthropic and OpenAI only.

Can I cap spend at the Ochre level too? Yes. The Ochre cap is independent of your provider limits and pauses AI replies the moment your monthly budget hits 100%. See Spend caps and alerts.

What's next

Once your key is in, walk through Choosing a model and try a few sample messages in the Playground before turning the AI on for live tickets.