Skip to main content

Decorators

With only a few lines of code, builders can easily deploy custom AI agents to the Naptha hub or their own local node.

Context​

Naptha supports a web of multi-agent systems that grow and evolve. This walkthrough explains how to decorate functions in order to quickly publish existing agents, allowing them to interact with others.

Prerequisites​

  • Python >= 3.10, <= 3.13
  • Poetry package manager
  • Naptha username & password
tip

Need help installing? Check out our detailed installation guide.

Step-by-step Walkthrough​

1. Setup​

Configure your .env file:

HUB_USERNAME=<your_naptha_username>
HUB_PASSWORD=<your_naptha_password>
HUB_URL=wss://hub.naptha.ai/rpc

NODE_URL=https://node.naptha.ai
info

These variables are used to connect to the Naptha network.

2. Add Naptha SDK​

Update your pyproject.toml file:

[tool.poetry.dependencies]
naptha-sdk = {git = "https://github.com/NapthaAI/naptha-sdk"}

If not using poetry, you can just add naptha-sdk to your requirements.txt file, then run:

pip install -r requirements.txt

3. Install dependencies:​

Execute via CLI:

poetry install

4. Import Naptha SDK​

Put this in your main Python file: <agent>.py

from naptha_sdk.client.naptha import agent as naptha_agent

We recommend importing as naptha_agent to avoid naming conflicts.

5. Decorate​

Label agent functions:

@naptha_agent("<agent_name>")
def <agent_function>(...):
    # agent logic goes here
    return <agent_output>

Replace <agent_name> with a unique identifier for your agent. Keep <agent_function> and <agent_output> the same, however you named them.

6. Convert (automatically)​

Process and package decorated functions:

poetry run python <agent>.py

This creates a folder named agent_pkgs, which contains your "Napthafied" agent functions. The SDK translates your code into a format compatible with other agents on Naptha.

7. Test​

cd agent_pkgs/<agent_name>
poetry install
poetry run python <agent_name>/run.py

If you're using an agent framework other than CrewAI, you will likely need to change the inputs dict, where tool_name is the name of the agent method that you would like to call, tool_input_type is the type (e.g. pydantic schema name) of the input for that call, and tool_input_value is a dict of the schema parameters and values.

8. Publish​

Enter this command:

naptha publish

This command publishes all agents in the agent_pkgs folder to the Naptha node specified in your .env file.

Verify the agent is working properly:

naptha run <agent_name>

Check for your expected output.

How does this work?​

Let's break it down:

  • @naptha_agent("<agent_name>") with a unique name for the agent. Below that line, agent functionality can be defined normally using various frameworks.

  • By running Python code that includes our decorator, agent functions are automatically processed and converted into Naptha-compatible packages.

  • Later, when you enter naptha publish via the CLI, those agent packages will be added to the Naptha node specified in your .env file.

Usage Examples​

  1. CrewAI
  2. Autogen
  3. Langchain
  4. LlamaIndex

Other Frameworks​

info

Support for additional agent frameworks is coming soon! Stay tuned for updates.

Feedback​

We welcome your feedback and contributions! Here's how you can help:

  • Report bugs and request features by creating issues on GitHub.
  • Share your example implementations and use cases on Discord.
  • Ask questions and join discussions on Discord.

Your input helps make Naptha better for everyone. We're actively expanding our examples and documentation based on community feedback.