Choosing BYOK vs Pooled inference mode
Flaxxa AI runs in two modes. Picking the right one depends on procurement, finance, and compliance — not on engineering preference.
Pooled mode (default)
Flaxxa uses pooled provider keys at wholesale rates. We pass through provider pricing with 0% markup; you pay Flaxxa a single bill for AI usage along with the rest of your subscription.
Best for: early-stage teams, teams without procurement-grade requirements, teams who want one bill instead of two, teams who want to A/B test providers without managing keys per provider.
BYOK metadata-only mode (Pro tier and above)
You upload provider keys (OpenAI, Anthropic, Google, etc.) into the Flaxxa key store. Provider calls go directly to the provider with your key; Flaxxa sees only metadata (request id, token count, latency, status code) and never the request body or response. Provider invoices you directly.
Best for: teams with strict data-residency requirements, fintech / health / regulated industries, teams that already have a provider contract at a discount, teams whose security team requires response data to never leave the provider.
The decision matrix
| Question | Answer | Mode |
|---|---|---|
| Will security require responses never touch a third party? | Yes | BYOK |
| Do you have a discounted contract with the provider already? | Yes | BYOK |
| Do you want one consolidated bill? | Yes | Pooled |
| Are you running multi-region with strict data residency? | Yes | BYOK (per-region keys) |
| Are you experimenting / pre-product-market-fit? | Yes | Pooled |
Mixed mode
You can run mixed: BYOK for the heavy / regulated workloads, Pooled for the long-tail of low-cost calls. The gateway routes per-feature based on the policy you set.
Switching modes later
You can switch modes at any time without re-deploying. The gateway URL stays the same; only the routing-policy target changes. New keys take effect from the next request onwards.
Was this article helpful?
