Skip to main content

What is Agent Foundry?

Agent Foundry provides a robust foundation of software baselines and packages that tackles the repeatable, complex aspects of building agentic systems while ensuring reliability, scalability, and security.

The Challenge​

Every team building production AI agent systems runs into the same set of problems:

  • Silos and Duplication - The same foundational infrastructure gets rebuilt on every project
  • Extended Timelines - Year-long prototyping efforts before a system is ready for production
  • Cost & Talent Constraints - Senior engineering talent repeatedly reinventing solved problems
  • Prototype Gap - POCs lack production-ready security, observability, and hardening

Agent Foundry solves this challenge. It accelerates delivery of immediately deployable solutions, reduces complexity and risk, and allows teams to focus on mission objectives.

The Solution​

Agent Foundry provides two types of baselines:

Shared Libraries (PyPI Packages)​

Reusable Python packages maintained by the Foundry team that teams pip install (or uv add) into their projects. Versioned, back-compat matters, and they live behind a stable API surface.

Examples: Foundry Agent Packages — DI container & protocols (foundry-agent-core), YAML config loader (foundry-agent-config), FastAPI middleware (foundry-agent-fastapi), and the Strands SDK adapter (foundry-strands-agent).

Starter Templates​

Repositories designed to be forked, renamed, and customized. You own the fork from the moment you clone it.

Examples: Strands Base Agent — fork-and-customize agent backend with auth, sessions, MCP tools, A2A, and observability wired up. The Temporal Workflow Worker and API starters (coming soon) follow the same model for workflow-orchestrated systems.

Key Principles​

Framework-Agnostic​

Build agents with any framework (AWS Strands, LangChain, custom). When agents expose the A2A protocol, they can communicate with each other and be orchestrated by workflow engines.

Configuration-First​

Agent behavior is defined through declarative YAML configuration, not code changes. Single configurations work across all environments through environment variable injection.

Production-Ready from Day One​

Every component is designed for production deployment from the start with authentication, observability, security, and scalability built in. Baselines are hardened against the DISA Application Security and Development (ASD) STIG and ship their per-baseline checklists so inheriting teams get a head start on ATO.

Fork-and-Own Model​

Teams fork specific tagged versions and build their mission-specific capabilities on top. Solutions with broader applicability flow back into Foundry.

What Foundry is NOT​

  • Not a hosted platform - Teams deploy and operate baselines themselves
  • Not mission-specific - Components are intentionally general-purpose
  • Not a drop-in solution - Engineering effort is required to customize for your use case

Next Steps​