Installation

Install the packages you need for the golden path.

Python packages

Create a project folder and virtual environment:

mkdir hello-agentflow
cd hello-agentflow
python -m venv .venv
source .venv/bin/activate

Install the core Python library and CLI:

pip install 10xscale-agentflow 10xscale-agentflow-cli

Verify the CLI is available:

agentflow version

The core library gives you StateGraph, AgentState, and Message. The CLI gives you commands like agentflow init, agentflow api, and agentflow play.

TypeScript client

If your app calls AgentFlow from TypeScript, install the client package:

npm install @10xscale/agentflow-client

You can do this in your application project or in a small Node.js test project.

The client package does not run your graph. It sends requests to a running AgentFlow API server.

Model provider keys

The first Python example in this guide does not call a model. When you switch to an LLM-backed agent, set the key for the provider you use:

export OPENAI_API_KEY="your-openai-key"

Use the environment variable that matches your model provider.

For this guide, provider keys are optional until you replace the demo node with an LLM-backed agent.

Contributor workspace

If you are working inside the AgentFlow repository instead of a new app, use the shared repository virtual environment:

cd /Users/shudipto/Projects/agentflow
source /Users/shudipto/Projects/agentflow/.venv/bin/activate

Then move into the package you are changing before running package-specific commands:

cd agentflow

Package roles

Package	Install when
10xscale-agentflow	You build Python graphs, agents, tools, state, checkpointers, or storage.
10xscale-agentflow-cli	You want agentflow init, agentflow api, and agentflow play.
@10xscale/agentflow-client	You call an AgentFlow API from TypeScript.

Next step

Build your first Python agent.