Why I built promptev on BYOK (and never stored a customer's data)
Bring-your-own-key looked like a commercial handicap. It turned out to be the architecture decision that makes enterprise conversations short.
When I was sketching promptev’s architecture, the conventional SaaS playbook said: sit between the customer and the model providers, mark up the tokens, own the data, own the relationship. Usage-based margin on every call. Very fundable. Very standard.
I went the other way — bring your own keys, and customer data never stored — and it’s the decision I’ve had to defend most and regretted least.
What BYOK actually means here
Customers plug in their own API keys — OpenAI, Anthropic, Google, Cohere, Azure, Bedrock, whatever mix they run. promptev orchestrates: agent definitions, the DRAFT framework, context packs, tool connections, approval flows, evaluation. Inference happens on their accounts, under their data-processing agreements, at their negotiated rates. Documents flow through the Context Engine for processing but the platform doesn’t retain them.
The platform is the conductor. It never becomes the vault.
The case against (I heard it plenty)
“You’re leaving margin on the table.” True — no token markup.
“Onboarding has more friction.” Also true. A key is one more step than a credit card, and I’ve watched people drop at that step.
“You can’t build features on aggregated customer data.” Correct, and permanently so.
All real costs. Here’s what they bought.
The security review that isn’t
Sell AI tooling to an enterprise and there is a meeting — you know the one — where security asks where prompts go, where documents rest, which subprocessor sees what, and what happens to their data when the contract ends.
With BYOK and zero retention, my part of that meeting is short. Inference: your keys, your provider, your existing DPA — a vendor you already approved. Data at rest: there isn’t any, here’s the architecture doc. I’ve watched that answer collapse procurement timelines from quarters to weeks. For a bootstrapped product without a compliance department, that’s not a nice-to-have. It’s the difference between selling and not.
Incentives you don’t have to apologize for
A platform that marks up tokens earns more when you burn more. A platform that holds your corpus gets stickier the harder leaving becomes. Neither incentive points at the customer’s interests, and enterprise buyers — who’ve been burned by both — can smell it.
promptev’s cost-aware execution exists because the margin isn’t mine: routing to cheaper models, caching, dedupe — features that shrink the customer’s bill make the product better, not the revenue smaller. It’s pleasant to build features whose incentives you’d happily explain to the buyer’s CFO.
Would I choose it again
For this product, this buyer, this era of AI procurement anxiety — yes, without hesitation. Model providers keep releasing; customers keep multi-homing; BYOK means promptev surfs that churn instead of fighting it. Multi-LLM routing across your own keys was a weird feature in 2024 and is table stakes now.
The playbook optimizes for owning the customer. I’d rather be easy to adopt and hard to distrust. Ask me again at scale — but so far, trust has compounded better than margin would have.