harness-sdk
View on GitHubBuild an agent harness and control it end-to-end. Open-source SDK for production AI agents in Python & TypeScript - any model, any cloud.
Strands Agents is an Apache-2.0 SDK for building and running AI agents in Python and TypeScript. It provides an in-process agent loop with tools, MCP, multi-agent patterns, memory, streaming, guardrails, tracing and evals, and is model-agnostic across Bedrock, Anthropic, OpenAI and Gemini.
Use Cases
building production AI agents in Python or TypeScriptcustom agent loops with turn limits, token budgets and cancellationmulti-agent orchestration patternsconnecting agents to MCP toolsmodel-agnostic provider switching (Bedrock, Anthropic, OpenAI, Gemini, Ollama)structured output from agentsstreaming agent responsesguardrails and steering handlers for self-correctionagent session and memory managementtracing and observability of the agent looprunning agent evaluationsdeclarative tool definitions with Zod typing
Built With
- Language
- Python
- Frameworks
- MCP · Amazon Bedrock · Anthropic · OpenAI · Gemini · Ollama · Zod · Astro/Starlight · pytest · Vitest
Tags
agents · agent-framework · multi-agent · mcp · sdk · llm · python · typescript · streaming · structured-output · guardrails · observability · evals · memory · model-agnostic · tools