Strands Harness SDK
AI agent SDK for Python and TypeScript that runs in your process. Apache 2.0 licensed, with Ollama, cloud model providers, MCP and lifecycle controls.
Strands Harness SDK is an open source toolkit for developers who want to build AI agents while controlling how they run. It executes inside your own application process, with no hosted control plane. The Apache 2.0 license covers the SDK, which supports Python and TypeScript.
You choose the model backend. Ollama support lets agents use a local LLM, while Amazon Bedrock, Anthropic, OpenAI and Google Gemini connect them to cloud models. The agent loop runs in your process in either case; cloud providers handle model requests outside it. You can change providers without rewriting the rest of the agent code.
The SDK handles repeated model calls and tool use, with controls for turn limits, token budgets and cancellation. Hooks let your application inspect, validate or redirect individual steps. Context management, memory and sessions support agents that need to retain information across interactions, while tracing records the agent's decisions.
Tools can come from your own functions or MCP servers. Structured output lets agents return data in a defined form, and streaming exposes responses as they're generated. The SDK also supports multi-agent patterns, guardrails and evaluation.
For a preassembled starting point, the related Strands harness supplies defaults for prompts, tools, memory and context management. It uses the same SDK, so developers can replace those defaults or work directly with the underlying components.
