bwaba.ai sits between your agents and any LLM — redacting PII, routing across providers, and metering tokens to the cent. Boring on purpose.
Most "AI gateways" stop at routing. bwaba is opinionated about the two things that actually keep you up at night — leaking customer data, and the bill.
Inline DLP rewrites names, emails, card numbers, tickets, internal IDs — anything you mark as sensitive — before the prompt leaves your perimeter. Restored on the way back so your agents never know.
OpenAI, Anthropic, Google, Mistral, Meta, xAI, Groq, DeepSeek, Cohere, Together, plus your own private endpoints. Swap models with a string. Route by latency, cost, region or capability — per request.
No seats, no minimums, no "starter plans." You pay the provider's token rate plus a flat 4% — billed by the cent, itemized by route, model, agent and tenant. Pause anytime; usage goes to zero.
Every prompt is parsed, tokenized for sensitive entities, and rewritten with stable placeholders. The model sees a sanitized version. Your code sees the original. The provider sees nothing you didn't approve.
Hi, I'm Jane Doe (employee #E-4419). Can you draft a reply to m.alvarez@northwind.co about invoice A-44912 for €18,420.50? My card ending in 4242 was charged twice.
Hi, I'm <NAME_01> (employee <EMP_ID_01>). Can you draft a reply to <EMAIL_01> about invoice <INV_01> for <AMOUNT_01>? My card ending in <CARD_01> was charged twice.
Every provider ships the day they ship. Use "model": "auto"
and we'll pick — or pin a specific one.
Spin up a key, point your SDK at api.bwaba.ai, and watch sensitive tokens disappear in real time. Free up to 1M tokens / month.