px0 is an open-source prompt infrastructure and LLM agent registry platform. It provides a unified control plane to version, update, execute, and govern prompt templates, multi-file agent skills, and structured tool schemas in production - completely decoupling prompt engineering and agentic logic from your core application deployment cycle.
- Decouple prompts from code using versioned Jinja templates with sandboxed rendering.
- Push instant, deploy-free prompt updates and rollbacks with visual diff tracking and granular team-level role-based access control.
- Run prompts natively with automated routing to OpenAI, Anthropic, DeepSeek, Gemini, and other providers.
- Stream responses using a unified server-sent events format and support dynamic, in-flight model overrides.
- Package and version complex multi-file agent logic and assets in structured workspaces.
- Promote skills through draft, stable, and live pipelines with absolute execution safety and zero behavioral drift.
- Model, validate, and govern structured tool schemas with full JSON Schema compliance.
- Expose metadata for LLM orchestration and perform single-click atomic schema updates.
Follow these steps to spin up the local services and send your first request. Refer to the Getting Started Guide for detailed verification and user registration instructions. For telemetry setup, metrics observation, and benchmarking details, refer to the Telemetry and Benchmarking Guide.
Clone the repository, configure the environment, and boot the background containers in detached mode.
cp .env.example .env
docker compose up -dThis starts the Go API server, PostgreSQL, Redis, and the OpenTelemetry observability stack.
Send a health check request to confirm the API server is active and listening on port 8000.
curl -i http://localhost:8000/v1/healthOpen the pre-configured observability dashboards in your browser.
- Grafana Dashboards: http://localhost:3000
- Prometheus Expression Browser: http://localhost:9090
px0 includes a built-in concurrent load testing utility located at cmd/loadtest/main.go. This script automatically handles transaction-safe database setup, concurrent execution, and lock-free metric collection.
Refer to the Benchmarking and Performance Guide for execution flags, metric details, and latency percentiles.
px0 ships an MCP (Model Context Protocol) server that exposes every documented API operation as an MCP tool, so AI assistants like Claude can manage prompts directly. Refer to the MCP Server Guide for startup instructions, per-client authentication, and testing with MCP Inspector.
You can find examples in the px0 examples repository. Get started with our SDK-based hello worlds:
This project is licensed under the MIT License.