Skip to main content

An Open Specification for Prompt Engineering

Define, test, and manage AI agent behavior in a framework-agnostic format—portable across any provider or runtime.

The Problem

AI agent development is fragmented. Each framework has its own format for prompts, tools, workflows, and test scenarios.

When you need to change frameworks or providers, you rebuild from scratch. Agent logic—now critical business logic—lacks engineering discipline.

The Solution

PromptPack provides a vendor-neutral JSON format for packaging AI agent behavior:

  • Framework-agnostic prompt definitions
  • Reusable tool specifications
  • Workflow and agent-loop orchestration
  • Version-controlled agent logic
  • Works with ANY runtime or provider

Key Capabilities

  • Agent Loops (v1.4): Terminal states, retry budgets, replayable artifacts
  • Workflows & Multi-Agent (v1.3): State-machine orchestration with A2A definitions
  • Multi-Prompt Architecture: Specialized prompts per scenario or workflow stage
  • Tool Integration: Define tools once, use across all prompts
  • Evals & Guardrails: Quality policy and safety travel with the pack
  • Complete Packaging: Everything in one portable JSON file

Stop Building the Same Agents Over and Over

Package your prompts, tools, workflows, and agent loops into a single deployable unit. Share proven agent behavior across teams, projects, and organizations. Build once, deploy everywhere.

Deploy Agents, Not Just Prompts

Everything your agent needs in one JSON file: specialized prompts, external tools, workflow state machines, safety guardrails, and shared resources. No more hunting for scattered configurations. One file = Complete agent.

Know Your Agent Works Before You Deploy

Built-in evals and testing metadata show which models work best, expected costs, and success rates. Deploy with confidence knowing your agent performs reliably. No more "works on my machine" surprises.