The self-hosted LLM gateway
for agencies & agent systems

What Promptopoly actually does for you

Six reasons teams run the gateway — each with the mechanism behind it.

• No agent can blow up the bill

  The budget is checked before every call, not invoiced afterwards. An agent stuck in a loop is stopped while it has still cost you nothing.

  Daily, weekly and monthly budgets per tenant and project · prepaid wallet with atomic debit · refund on stream abort

• One key instead of five provider accounts

  A single pk_ key for OpenAI, Anthropic and Google (Gemini). A new model arrives through configuration, not through a deployment.

  OpenAI-compatible API · native /v1/messages endpoint · streaming and tool calling

• Provider keys never sit in your code again

  The real provider keys stay in the KeyVault. Applications, agents and outside contractors only ever see the pk_ key — and that one can be revoked on its own.

  AES-256-GCM encrypted · plaintext only transient in RAM · rotation and fingerprinting included

• Every call stays provable — months later

  Who used which model, when and for what, is recorded without gaps in a hash chain. That answers audit and compliance questions without rebuilding anything in a spreadsheet.

  HMAC hash chain · export as CSV or JSON · fixed deletion periods, 7-day response retention

• Usage agencies can actually bill on

  Costs accrue separately per project and per client. The log becomes the line item on the client invoice — instead of one lump sum nobody can split.

  Reporting per tenant and project · model, token count and cost for every call

• A provider outage no longer costs you a night

  If a provider goes down, a fallback model takes over automatically. The failed attempt is credited back to your balance.

  Curated model aliases · automatic failover with refund · made in Germany

And the context your agents work with lives in the same account: searchable, linked, reachable over the same API — the Brain Repository.