AppNewofficial publisher

Strands Agents

App : A product you use directly

AWS open source SDK for writing an AI agent in a few lines of Python or TypeScript: the model drives the loop, you declare the tools.

The shortest path from a script to a production agent on AWS, without lock-in: it supports Bedrock, Anthropic, OpenAI, Gemini, or Ollama. Designed to be deployed on Bedrock AgentCore.

Official site GitHub Docs

Overall score

7.0 /10

Security: one caveat, see the card

Utility

8.0

Reliability

6.0

Security

One caveat

Ease

5 min · Technical

Security reflects where data goes, requested permissions and incidents on record at the verification date. It is not a guarantee of zero risk: check the deployment mode and the connections you enable.

What it does, concretely

Building an AI agent capable of chaining autonomous actions typically requires manually managing the reasoning loop, tool calls, and context handling. Strands Agents is an open source SDK (Apache 2.0) published by AWS that handles this mechanics, letting you declare your tools and describe your goal.

In practice, you annotate a Python function with @tool or declare it in TypeScript, instantiate an Agent by passing a model and a list of tools, then invoke it with a natural language instruction. The SDK manages the loop: the model decides which tools to call, in what order, until the result is reached. Strands Harness adds a ready-to-use layer with optimized defaults for memory, sessions, and context management.

Strands Agents targets Python or TypeScript developers who want to move from a prototype to a production agent on AWS, while retaining the freedom to choose their model provider (Bedrock, Anthropic, OpenAI, Google, Ollama). Companies such as Smartsheet, Swisscom, and Landchecker use it in production.

It is not suited for people without programming experience: there is no graphical interface, everything is configured in code. Inference and hosting costs remain the responsibility of the user, depending on the chosen model provider and infrastructure.

What it offers

·Tool declaration with a single annotation (@tool) from any Python or TypeScript function, with the schema automatically generated from type hints and docstrings.
·Compatibility with multiple model providers: Bedrock, Anthropic, OpenAI, Google, and Ollama, interchangeable without rewriting the rest of the code.
·Strands Harness provides a pre-built agent with optimized defaults for memory, sessions, and context management, usable as a CLI or a library.
·Strands Shell embeds a Bourne-compatible shell that runs inside the process without fork or exec, with file, URL, and credential access explicitly declared.
·Connection to MCP (Model Context Protocol) servers to plug in data sources or external tools without additional code.
·Strands Evals offers more than 25 built-in evaluators to score response quality, tool selection, trajectory, and safety before shipping (Python only).
·Strands Labs provides experimental projects including control of physical or simulated robots via natural language, with more than 70 robot models available.
·Support for multi-agent patterns (Agent-as-Tool, Swarm) and native integration with Amazon Bedrock AgentCore, Bedrock Guardrails, and OpenTelemetry.

Our tips

Use Strands Harness (create_harness) to get started quickly with optimized defaults, then move down to the Harness SDK only if you need to control every component of the loop.
With Strands Shell, explicitly declare directories (binds) and the network allowlist before running the agent: anything not declared is invisible to the model, limiting the exposure surface.
To evaluate an agent before deploying it, run Strands Evals with test cases covering the full trajectory (not just the final answer) to detect the steps where reasoning goes wrong.
Add the interventions parameter in create_harness to require human confirmation before sensitive actions (refunds, deletions), without modifying the tool logic.
Strands Labs projects (robots, benchmarks) evolve faster than the core SDK and may introduce frequent changes: reserve them for prototypes or test environments.

Watch out

Strands Evals is available in Python only: TypeScript projects cannot use the built-in evaluation module.
Strands Labs projects are explicitly marked as experimental and may change frequently, making them unsuitable for stable production use.

In practice

Price: Free
Commercial use: Commercial use: yes
Available on: cli
Licence: Apache-2.0
FR interface: No

Security and data

Rating: Security: one caveat, see the card
Your data: Local or cloud depending on use
Maintenance: Active

Links and repository checked on 18 October 2026 · added on 27 September 2026

How to install Strands Agents

1. Verify that Python 3.10+ or Node.js is installed on your machine. 2. Install the SDK: pip install strands-agents (Python) or npm install @strands-agents/harness (TypeScript). 3. Configure credentials for your chosen model provider (environment variable or AWS, Anthropic, OpenAI configuration file, etc.). 4. Declare your tools with @tool (Python) or tool() (TypeScript), limiting permissions to the strict minimum. 5. Instantiate the agent with Agent() or create_harness() and invoke it with your instruction. 6. Before any deployment, run Strands Evals (Python) to validate the agent trajectory on your test cases.

npm install @strands-agents/harness

Official documentation ↗

In the same category

Whole category Automation & agents →