Skip to main content

Initialize a project

agentflow init scaffolds the minimum files needed to run an agent behind the API: a configuration file, a graph module, and optional project structure files.

Prerequisites​

You need AgentFlow CLI and Python 3.11 or later.

pip install 10xscale-agentflow-cli
python --version  # Verify Python 3.11+

If you want to use the starter agent with Gemini, also install the client:

pip install google-generativeai

Create a project directory​

Create and navigate to your agent project directory:

mkdir my-agent-project
cd my-agent-project
python -m venv venv
source venv/bin/activate  # On Windows: venv\\Scripts\\activate

Run init​

Initialize the project scaffolding:

agentflow init

This creates:

agentflow.json
graph/
  __init__.py
  react.py
skills/
  agent-skills/
    SKILL.md
    references/
      skill-concepts.md
      agentflow.md
      claude.md
      codex.md
      github.md
.env

To initialize in a different directory without changing folders:

agentflow init --path ./my-agent-project

What each file does​

agentflow.json​

This is the core server configuration file. The default content:

{
  "agent": "graph.react:app",
  "thread_name_generator": null,
  "env": ".env",
  "auth": null
}

Field explanation:

  • agent (required) — the import path to your compiled graph, expressed as module:attribute. The loader imports the module and retrieves the attribute (usually a CompiledGraph instance). In this case, it imports the graph.react module and retrieves the app variable.
  • env (optional) — path to a .env file containing environment variables. These are loaded with python-dotenv before the graph module is imported, so you can reference them in your graph initialization.
  • thread_name_generator (optional) — import path to a thread name generator. When set, the API uses it to create human-readable thread names (e.g., thoughtful-dialogue) automatically when threads are created.
  • auth (optional) — authentication configuration. Set to null for no auth, "jwt" for JWT bearer tokens, or a custom auth backend configuration.

graph/react.py​

A starter ReAct (Reasoning and Acting) agent. This is a reference implementation using Google's Gemini model. The example shows:

  • How to define a simple state type with message history
  • How to create a graph with a single agent node
  • How to compile the graph (which enables checkpointing and real-time updates)
  • How to prepare the graph as a FastAPI route (via graph.invoke() and graph.stream())
  • How to discover the generated skills/ directory and expose a set_skill tool

You must keep an app variable defined that holds the compiled graph. Replace the graph logic with your own while preserving this export.

skills/​

Starter agent skill library for Claude, AgentFlow agents, Codex, and GitHub workflows. It follows a compact main-file plus references structure:

  • skills/agent-skills/SKILL.md — main skill entry point and reference map
  • skills/agent-skills/references/skill-concepts.md — skill layout, frontmatter, resources, and validation
  • skills/agent-skills/references/agentflow.md — AgentFlow graph and SkillConfig patterns
  • skills/agent-skills/references/claude.md — Claude-facing repository guidance
  • skills/agent-skills/references/codex.md — Codex skill authoring workflow
  • skills/agent-skills/references/github.md — GitHub MCP and repository automation workflow

The generated graph/react.py discovers this folder and adds a set_skill tool so the model can load agent-skills and request exact reference files during a run.

.env​

Stored environment variables for local development. The default is empty. Add API keys and connection strings here:

# Example
GOOGLE_API_KEY=your_key_here
REDIS_URL=redis://localhost:6379/0

The agentflow.json field env tells the server which .env file to load.

Overwrite existing files​

If you have already initialized a project and want to regenerate the starter files:

agentflow init --force

This overwrites agentflow.json, graph/__init__.py, graph/react.py, and generated files under skills/. Use --force carefully—it will replace existing code and skill instructions.

Initialize a production project​

The --prod flag scaffolds additional files for a professional project layout:

agentflow init --prod

This creates:

  • pyproject.toml — Python package definition (allows pip install -e .)
  • .pre-commit-config.yaml — Git hooks for code quality checks (linting, formatting)
  • Standard project structure with src/, tests/, and docs/ directories

Use --prod when you are setting up a project that will be deployed or shared with a team.

Verify the initialization​

After running init, verify the files were created:

ls -la
cat agentflow.json
cat graph/react.py

You should see valid Python and JSON files with no syntax errors.

Test the starter agent​

Before modifying the starter agent, test that it runs:

# Set a Gemini API key if you want to test the starter
export GOOGLE_API_KEY=your_key_here

# Start the server
agentflow api --host 127.0.0.1 --port 8000

In another terminal, test the agent:

curl -X POST http://127.0.0.1:8000/v1/graph/invoke \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{"role": "user", "content": "What is 2+2?"}],
    "config": {"thread_id": "test-1"}
  }'

Expected response: A JSON object with messages, including the agent's response.

If this fails, check:

  • Is the server running? (curl http://127.0.0.1:8000/ping should return {"success": true, "data": "pong"})
  • Is GOOGLE_API_KEY set? (The starter example requires it)
  • Are there errors in the server logs?

Next steps​

  1. Replace the agent — Edit graph/react.py with your custom graph logic. See concepts for how graphs work.
  2. Add tools — Give your agent callable functions via ToolNode. See add a tool guide.
  3. Add persistence — Configure a checkpointer in agentflow.json so conversations persist across server restarts. See configure agentflow.json.
  4. Start the server — Run agentflow api to expose your agent over HTTP.
  5. Test with the playground — Use agentflow play to test interactively in a browser.

Troubleshooting​

Error: "ModuleNotFoundError: No module named 'agentflow_cli'"

  • Install the CLI: pip install 10xscale-agentflow-cli

Error: "No such file or directory: agentflow.json"

  • Make sure you run agentflow init from the project root, or use --path to specify the directory.

Error: "Overwrite existing files?" but you did not pass --force

  • The CLI prompts interactively. Type yes or pass --force to skip the prompt.

The generated graph/react.py has syntax errors

  • This is unusual. Try deleting the directory and re-running agentflow init from scratch.
Last updated on