Skip to main content

Overview

Status: Available Version

Base agent runtime for deploying AI agents built on the AWS Strands Agents SDK.

Source: github.com/boozallen/strands-base-agent

Strands Base Agent is a baseline starter repo from the Agent Foundry team. It accelerates AI agent delivery for client engagements and team POCs by providing a FastAPI-based HTTP server wired to the AWS Strands Agents SDK, with shared infrastructure abstracted into published foundry-agent-* packages. Teams fork this repo into their own environment, configure their model and system prompt in config.yaml, add domain-specific tools, and deploy.

Key features​

  • YAML-first configuration — All agent settings live in config.yaml with environment variable overrides using double-underscore nesting
  • Multi-provider LLM support — AWS Bedrock, Ollama, and LlamaCpp via the Strands Agents SDK
  • A2A protocol server — Built-in Agent-to-Agent communication with automatic agent card discovery at /.well-known/agent-card.json
  • MCP server integration — Connect external tools via Model Context Protocol with declarative config
  • OpenTelemetry observability — OTLP export for tracing and metrics out of the box

Architecture​

The application is a FastAPI server (strands_base_agent/server.py) that exposes REST endpoints for query, streaming, and chat history. A composition root (application/factory.py) wires together foundry-agent-config (Pydantic-based configuration), foundry-agent-core (agent lifecycle), foundry-agent-fastapi (route adapters), and foundry-strands-agent (Strands SDK integration). Domain-specific tools are loaded from the tools/ directory at startup. Session state is managed via pluggable backends (file-based or S3).

Security posture​

Strands Base Agent inherits its security baseline from the upstream foundry-agent-* packages — static analysis (Ruff + bandit) runs in both just lint and CI, and DISA ASD STIG checklists are tracked at the package level. See Foundry Agent Packages → Security posture for the shared checklist surface.

Where to go next​

Extending this baseline​

Planning to add a tool, swap a backend, or harden this fork for an engagement? See Spec-Driven Extensions with OpenSpec for the propose-then-implement workflow we recommend — especially useful when a coding agent will do the work.