The harness-sdk repository is a monorepo containing the Strands Agents SDK, an open-source framework for building and deploying production AI agents in Python and TypeScript. The project takes a model-driven approach to agent development, allowing developers to create agents with minimal code while maintaining full control over execution and behavior. The repository is organized into distinct directories: the Python SDK under strands-py, the TypeScript SDK under strands-ts, a developer CLI called strandly, documentation infrastructure in the site directory, and governance materials in the team directory.
The core value proposition centers on flexibility and control. The framework supports any large language model and any cloud provider, with first-class integrations for Amazon Bedrock, Anthropic, OpenAI, and Gemini, plus additional providers and custom implementations. Developers can swap between backends without changing their application code. Built-in features include context management, execution limits, observability, streaming support, structured output, MCP protocol support, and multi-agent patterns. The agent loop traces every decision by default, and developers can intercept any step through hooks for logging, validation, or redirection.
The Python SDK requires Python 3.10 or later and is distributed via PyPI as the strands-agents package. The TypeScript SDK requires Node.js 20 and is available on npm as @strands-agents/sdk. Both SDKs default to Amazon Bedrock as the model provider, though the documentation covers configuration for alternative providers including Ollama for local development. The TypeScript implementation includes Zod-typed tools and structured output capabilities alongside the core agent loop functionality.
GitGenius activity data reveals substantial development momentum and community engagement. The most active labels are enhancement with 602 items, bug with 370 items, and python with 223 items, reflecting ongoing feature development and maintenance.
The repository emphasizes production readiness through guardrails that catch mistakes before execution and steering handlers that allow agents to self-correct rather than fail silently. The documentation site, built with Astro and Starlight, provides comprehensive guidance including a quickstart guide, agent loop concepts, examples, API references for both languages, and a production deployment guide. The project maintains a Discord community for user engagement and team interaction.
Development is coordinated from the repository root with git operations, while each package maintains its own toolchain and release cycle. The framework is designed to scale from simple conversational assistants to complex autonomous workflows, supporting both local development and production deployment scenarios.
