Own product / open source
PromptZero: AI prompt management
Platform to create, version, run and compare prompts across OpenAI, Anthropic, Gemini and OpenRouter, with cost and latency analytics and A/B experiments.
- Case
- prompt zero
- Code
- Open source
- Stack
- Next.js 16 / React 19 / NestJS 11
- 4
- AI providers
- 15+
- supported models
- 3
- interface languages
- 100%
- of routes in Swagger
§ Overview
Teams using AI keep prompts in docs, spreadsheets and scattered through code, with no history, no visible cost and locked into one provider. PromptZero covers the whole lifecycle: every edit becomes a version, runs stream back live, A/B experiments pick the best variation with statistics, and the dashboard shows usage, cost and latency per model. Underneath it is a real production product: NestJS with Clean Architecture, JWT with refresh token rotation, encrypted API keys, full observability with metrics, logs and traces, unit and E2E tests, CI/CD and automated deploys.
- Next.js 16
- React 19
- NestJS 11
- Prisma
- PostgreSQL
- TypeScript
- Redis
- Tailwind CSS v4
- OpenTelemetry
- Grafana
- Docker
- Playwright
B/01
Challenges
- Prompts scattered across docs, spreadsheets and code, with no history
- No visibility on cost and latency per model
- Comparing variations without an objective criterion
- Lock-in to a single AI provider
B/02
Solutions
- Automatic versioning with soft delete and full history
- Provider-agnostic layer with retry, backoff and circuit breaker
- A/B experiments with traffic split and statistical analysis
- Observability with Prometheus, Loki, Tempo and Grafana
B/03
Features
- Response streaming via Server-Sent Events
- Templates with typed variables
- Workspaces, categories and favorites
- Public explore and prompt forking
- JWT with refresh token in an httpOnly cookie and AES-256-GCM on keys
- UI in Portuguese, English and Spanish
§ Inside, screen by screen
07 screensT/04
Editor and runner
Edit, version and run in one place. The editor accepts template variables, and the runner lets you tune temperature, top P, top K and token limits before streaming.
- Variables like {{company}} and {{main_pain}}
- Advanced model parameters
- Every save creates a new version
PromptZero screens running locally with the project's own demo data.
§ And there is more
+01
A/B experiments
Two variants side by side, configurable traffic split, voting and results with win rate and confidence.
+02
Public explore
Public prompts browsable without login, filtered by language, model and popularity, and forkable into your account.
+03
Observability
Metrics in Prometheus, structured logs in Loki and OpenTelemetry traces in Tempo, all correlated in Grafana.
+04
Quality
Jest and Vitest for unit tests, Supertest and Playwright for E2E, zero lint errors and strict TypeScript.
+05
Deploy
Docker Compose to run everything, GitHub Actions on every push and automated Railway deploys with migrations.
+06
Security
Per-route rate limiting, AES-256-GCM encryption of API keys and OWASP practices.






