PromptRails
View as Markdown

Introduction

PromptRails is an AI agent orchestration platform for building, deploying, and monitoring LLM-powered applications.

What is PromptRails?

PromptRails is a platform for building, deploying, and monitoring AI agents. It provides a complete infrastructure layer between your application and LLM providers, giving you full control over prompt engineering, agent orchestration, observability, and cost management.

Whether you are prototyping a single chatbot or running a fleet of production agents across multiple LLM providers, PromptRails provides the tools to manage the entire lifecycle.

Key Features

  • Agent Orchestration -- Build agents using five distinct execution strategies: simple, chain, multi-agent, workflow, and composite. Compose complex AI pipelines from reusable building blocks.

  • Prompt Management -- Version-controlled prompts with Jinja2 templating, input/output schemas, model assignment, and caching. Promote and roll back versions without code changes.

  • Multi-Provider LLM Support -- Connect to OpenAI, Anthropic, Google Gemini, DeepSeek, Fireworks, xAI, and OpenRouter through a unified credential system.

  • Tracing and Observability -- OpenTelemetry-style distributed tracing with 14 span kinds. Track every LLM call, tool invocation, guardrail evaluation, and data source query with full cost and latency breakdowns.

  • Guardrails -- 14 built-in scanner types for input and output validation including toxicity detection, PII filtering, prompt injection prevention, and more. Configure block, redact, or log actions per scanner.

  • MCP Tools -- First-class Model Context Protocol support. Connect external APIs, data sources, built-in functions, and remote MCP servers as tools for your agents.

  • Data Sources -- Query PostgreSQL, MySQL, BigQuery, Snowflake, Redshift, MSSQL, ClickHouse, or static files directly from your agents using versioned, parameterized query templates.

  • Memory System -- Five memory types (conversation, fact, procedure, episodic, semantic) with vector embedding support and semantic search for context-aware agents.

  • Scoring and Evaluation -- Score executions and spans with numeric, categorical, or boolean metrics. Support for manual scoring, API-based scoring, and LLM judge automated evaluation.

  • Human-in-the-Loop Approvals -- Pause agent execution at configurable checkpoints and require human approval before continuing. Integrate approval workflows via webhooks.

  • Cost Tracking -- Automatic per-execution and per-span cost calculation across all LLM providers. Workspace-wide cost summaries and per-agent cost analysis.

  • Agent UI Deployments -- Build and deploy interactive dashboards backed by your agents, prompts, and data sources. Multi-page layouts with a 12-column grid system and optional PIN protection.

  • A2A Protocol -- Agent-to-Agent communication via Google's A2A protocol with agent cards, JSON-RPC messaging, and task lifecycle management.

  • SDKs and CLI -- Official Python and JavaScript/TypeScript SDKs, a full-featured CLI, and an MCP server for IDE integration with Claude Desktop, Cursor, and Windsurf.

How It Works

PromptRails handles the full lifecycle of AI agents — from development to production. The platform provides:

  • A web dashboard for building and managing agents, prompts, and data sources
  • REST APIs for programmatic access to all platform features
  • Real-time tracing with detailed execution breakdowns
  • Secure credential storage with encrypted secrets

All credentials are encrypted at rest and never exposed in API responses.

Getting Started

The fastest way to start using PromptRails is to install one of the SDKs and execute your first agent:

NextQuickstart