Proxy Mode or Your Own API Key: Which Should You Choose?

IrisAI can reach an AI provider two ways, and the choice is less about features than about who you’d rather have a billing relationship with.

Worth saying up front: this isn’t a paid tier versus a free one. Chat, citations, the answerable gate and the full retrieval pipeline are in both. The differences are in the surrounding service — usage tracking, balance alerts and the account dashboard are Proxy-side — and in one technical detail worth knowing before you choose.

Semantic search needs an embedding model. Proxy provides one. In BYO, an OpenAI or Google key provides one too — but a key that only covers chat, such as an Anthropic-only setup, leaves IrisAI falling back to local keyword-based matching. Chat still works; semantic matching doesn’t. If you’re going BYO with Claude, add an OpenAI or Google key for embeddings under Knowledge Base.

Proxy mode

Your site talks to IrisAI, and IrisAI talks to the AI providers. You buy credits; we handle the provider accounts.

What you get

• One-click setup — no provider account, no API key to create or paste
• Models from several providers behind a single connection, switchable from a dropdown
• Usage tracking, rate limiting and balance alerts built in
• Credits that don’t expire, on up to five domains per account

The trade-off

Your requests pass through infrastructure you don’t operate, and you’re buying credits rather than paying a provider at cost. In exchange, you never manage a provider account or think about which model is deprecated this quarter.

BYO mode

Your server talks to your chosen AI provider directly, using your own key. Nothing routes through us.

What you get

• Provider pricing at cost, with no intermediary
• Any model your key can reach, including ones not in our dropdown
• A direct data-processing relationship with the provider, which some compliance reviews prefer
• No per-domain limit

The trade-off

You own the operational work: creating the account, managing the key, watching for deprecations, handling rate limits, and reading provider invoices. There’s no consolidated usage dashboard, because your usage lives in the provider’s console.

One thing to know about scanned PDFs

OCR for scanned documents runs through the IrisAI proxy endpoint regardless of mode. If you’re in BYO mode and want to index scanned PDFs, you’ll still need a proxy key for that path. Text-based PDFs are parsed locally and need nothing extra.

A simple way to decide

Choose Proxy if you want it working today, you’d rather have one bill and a usage dashboard, or you’re setting this up for a client who won’t manage an API key.

Choose BYO if you already have a provider account, you want cost price at volume, or your compliance position requires a direct relationship with the AI vendor.

You can switch later, and your settings and content are untouched — but check the embedding model before you do. Different providers produce vectors of different sizes, and an index built with one can’t be searched with another. Moving between Proxy and BYO-with-an-OpenAI-key keeps the same embedding model, so the index carries over. Switching to Google embeddings, or to a setup with no embedding key at all, does not.

If it happens, the plugin tells you plainly: an admin notice asks you to clear the vector data and run a full re-index. Nothing is lost except the time and credits to rebuild it — worth planning for rather than discovering.

---

Side by side

	Proxy	Bring your own key
Setup	One click, no signup	Create a provider account first
Billing	300,000 tokens free, then $4/mo or a $5 top-up	Direct from your provider
Models	OpenAI, Anthropic, Google in one dropdown	Anything your key reaches
Domains	Up to 5 per account	Unlimited
Usage dashboard	Included	Provider console
Semantic search	Included	Needs an OpenAI or Google key
Scanned-PDF OCR	Included	Still needs a proxy key