Made in Germany · GDPR-Compliant
The self-hosted LLM gateway
for agencies & agent systems
Hard budget caps before every call, one pk_ key for all providers, OpenAI-compatible and Anthropic-native — with a tamper-proof audit trail. Plus: the prompt platform for teams.
Why Promptopoly
One gateway, three guardrails
-
01 · KeyOne key for every provider
-
02 · CapHard budget caps before every call
-
03 · AuditTamper-proof audit trail
What Promptopoly is used for
- As the engine behind agent systems. Autonomous agents call models through the proxy. The budget is checked before every call — an agent stuck in a loop can no longer blow up the bill.
- As one key for every provider. A single
pk_key instead of separate provider accounts. OpenAI-compatible, Anthropic-native; switch models without touching code. - As a cost brake during development. Experiments run against a hard limit instead of an open credit card.
- As a prompt library for teams. 364 tested prompts across 27 categories sit in the gallery and clone into your own repository in one click — instead of hiding in Notion, Slack and chat history.
- As an audit trail for compliance. Every call is recorded with model, token count and cost in a tamper-evident audit chain.
- As a billing basis for agencies. Track and pass on usage per project or per client.
The Problem
Your prompts live everywhere — except in one place.
You never know which version is in production. Your team reinvents the wheel every month. AI subscription costs rise — with no plan.
Proxy Gateway
Security and control for every AI call.
Hard budget limits, encrypted provider keys, firewall rules and tamper-proof audit — before a request leaves the gateway.
Budgets & Wallet
Daily, weekly or monthly budgets per tenant and project. Prepaid wallet with atomic debit and refund on stream abort — hard cost caps before every call.
KeyVault
Provider keys stored AES-256-GCM encrypted. Plaintext exists only transiently in RAM. Key rotation and fingerprinting included.
Firewall
IP allowlists, model whitelists, capability checks and blocklist regex on payload. Forbidden calls block before budget reservation.
Streaming & Tool-Calling
Server-Sent-Events streaming and tool calls (finish_reason: tool_calls) over the OpenAI-compatible API — plus a native /v1/messages endpoint for Anthropic. Built for agents.
Audit trail
Every request is logged in a tamper-proof HMAC hash chain. Export as CSV or JSON. GDPR-compliant deletion after the retention period (7-day response retention).
Multi-provider
OpenAI, Anthropic, OpenRouter — via a unified OpenAI-compatible API. Curated model aliases and automatic failover with refund.
Plus: the prompt platform
From experiment to production — in one tool.
Promptopoly combines prompt management, multi-model testing, and team collaboration. Everything you need to run professional prompt engineering. Optional: sync with AnnexEdit for macOS.
Multi-Model Playground
Send the same prompt to GPT-4o, Claude 4 and Gemini 2.5 — compare results side-by-side. The aha moment that alone converts visitors.
Prompt Gallery
350+ curated, tested prompts in 27 categories. For marketing, code, support, legal and creative. Copy, adapt, deploy.
Prompt-as-Form
Any prompt with variables becomes a shareable form in seconds. Non-technical colleagues fill in and get results — without seeing the prompt.
Workflow Builder
Build visual prompt chains: Prompt → Parse Output → Condition → Next Prompt → Tool Call. Conditional branching, loops, parallel execution.
Git Integration
Version prompts like code: GitHub, GitLab, Gitea. Changes as PRs, code reviews for prompts, CI/CD pipeline for prompt deployment.
Team Workspace
Share repositories with Read, Write, or Admin roles. Soft-locks prevent silent conflicts. Audit trail shows who changed what.
AI gateway for agents
OpenAI-compatible proxy with hard budgets, streaming, tool calling and a native Anthropic endpoint. Provider keys stay in the KeyVault — your agent only sees pk_. → View gateway
Prompt Gallery
Prompts you can use in 30 seconds.
Every prompt is tested, categorized and ready to run.
Comparison
With vs. without Promptopoly
| Topic | Without Promptopoly | With Promptopoly |
|---|---|---|
| Starting point | 20 min staring at empty chat | Verified prompt in 30 seconds |
| Quality | Inconsistent, depends on daily form | Proven structure, reusable results |
| Team | Everyone uses different prompts | Shared standards, shared playbooks |
| Cost | Expensive AI subs without plan | €8–15/month, structured, controlled |
| Testing | One model, one attempt | Side-by-side against 3+ models |
| Versioning | Copy-paste in Google Docs | Git integration, PRs, reviews |
Questions and answers
Frequently asked questions
What is Promptopoly?
Promptopoly is a self-hosted LLM gateway. Calls to language models run through a single pk_ key instead of separate provider accounts. The budget is checked before every call; once it is exhausted the call is stopped rather than executed and billed.
How does Promptopoly stop an agent from blowing up the bill?
The cap applies before the call, not at billing time. An agent stuck in a loop hits a hard limit and is stopped. Every call is logged with model, token count and cost, so usage can be traced per project.
Which providers and models are supported?
The API is OpenAI-compatible and Anthropic-native. Switching models requires no code change, because your application keeps talking to the same interface.
How many prompts are in the gallery?
364 tested prompts across 27 categories, including content, code, support and legal. Browse them for free and clone them into your own repository in one click.
Is Promptopoly suitable as an engine for agent systems?
Yes, that is one of the main use cases. Autonomous agents call models through the proxy and get budget limits, unified access and logging without having to implement any of it themselves.
Where is the data processed?
Operations run in Germany and are designed to be GDPR-compliant.
What does Promptopoly cost?
There is a free tier. Current pricing is on the pricing page.
Start for free
Start with a free account. No credit card. No commitment. Just better prompts from day one.