Pragma: Build Cross-Harness AI Expert Teams & Reusable Assets

Pragma is an innovative, code-public desktop application designed to move beyond traditional, rigid AI workflows. Instead of merely chaining steps together, Pragma focuses on orchestrating diverse, cross-harness AI Expert Teams. It seamlessly integrates various agent harnesses, models, tools, context sources, and human decisions, allowing them to collaborate while distilling their interactions into exportable AI assets.

Key Features

• Cross-Harness Expert Collaboration: Break the silos between different AI agents and tools. Pragma ensures a smooth transition of context, tasks, and responsibilities across diverse specialists. It acts as the ultimate conductor, making your existing models and harnesses work together cohesively without losing critical artifacts or decisions.

• Dynamic Memory & Asset Accumulation: AI interactions shouldn't be ephemeral. As your Expert Teams operate, Pragma captures the context through a dynamic memory pipeline. Useful experiences, decisions, and facts are processed through controlled policies and promoted into stable, reusable knowledge or customized Skills.

How Pragma Works

Users can easily define, orchestrate, and manage specialized Experts and ExpertTeams by connecting to cloud model providers or local runtimes. As these teams tackle complex tasks, their accumulated experience turns into institutional knowledge, continuously leveling up your AI capabilities.

# Example: Running context operations without model credentials
pnpm --filter @pragma/examples example:context

# Example: Using a local agent CLI
dpnpm --filter @pragma/examples example:runtime-codex

Exportable AI Assets

Pragma treats your expert configurations, team structures, and refined context as portable assets. It supports versioning, exporting, and seamless sharing of these AI setups under a unified governance structure. Your innovative ways of using AI become tangible assets you can easily distribute to your team or the community.

Private Knowledge Management

All context is securely managed via a ContextStore, funneling multi-agent events into a central Memory Pipeline. This architecture enables automated, policy-driven distillation of raw interactions into structured, private knowledge repositories, ensuring that your data works for you over the long term.

Practical Applications

Pragma is a powerhouse for complex projects requiring multi-disciplinary AI collaboration. For example:

• AI-Native Delivery Missions: Assign specialized roles (e.g., UI Designer, Code Reviewer, Architect) to dedicated Experts across different harnesses. Pragma ensures they share a unified context and hand off artifacts flawlessly, making it the perfect tool for managing complex, multi-stage project delivery.

Integration Options

Users can choose from various integration paths tailored to their needs:

• Desktop Application: Your central hub for team configuration, mission oversight, and asset import/export.
• Pragma DSL: A descriptive language with built-in version control to define Expert capabilities, team structures, and memory policies.
• TypeScript Packages: Embed Pragma's core orchestration and memory engines directly into your own existing applications.

Current Status and Documentation

Pragma is currently in a preview stage, available for macOS, with additional support planned for Windows and Linux environments. Up-to-date documentation provides guidance on usage, agent architecture, context routing, and integration techniques.

Visit the documentation for comprehensive guides and support for implementing Pragma in your projects.