Build with ADK - SambaNova Integration Guide

The Agent Development Kit (ADK) is a modular, model-agnostic framework developed by Google for building AI agents. It supports workflows, tool usage, multimodal input/output, and evaluation — making it ideal for flexible, interactive agents.

Multimodal input

Understand and process images alongside text in a single agent workflow.

Tool usage

Call external tools and APIs at any step of your agent’s reasoning loop.

Visual to code

Turn a visual mockup into a working web page automatically.

Model-agnostic

Swap LLM providers without rewriting your agent logic.

Prerequisites

A SambaCloud account and API key
Python 3.9 or later (python --version to confirm)

Setup

Create and activate a virtual environment

python -m venv .venv
source .venv/bin/activate

python -m venv .venv
.venv\Scripts\Activate.ps1

Install the required libraries

pip install google-adk   # Core SDK for building agents
pip install litellm      # Adapter to connect ADK with SambaCloud

litellm acts as a proxy between ADK and SambaCloud’s OpenAI-compatible API — no custom provider code needed.

Set your SambaCloud API key

export SAMBANOVA_API_KEY=your-sambanova-api-key

Never commit your API key to source control. Add it to .gitignore if you use a .env file.

Configure the model

Use ADK’s LiteLlm wrapper to point your agent at SambaCloud. LiteLLM reads SAMBANOVA_API_KEY from the environment automatically.

from google.adk.agents import Agent
from google.adk.models.lite_llm import LiteLlm

agent = Agent(
    name="my_agent",
    model=LiteLlm(model="sambanova/Meta-Llama-3.3-70B-Instruct"),
    description="A helpful assistant powered by SambaCloud.",
    instruction="You are a helpful assistant.",
)

Replace Meta-Llama-3.3-70B-Instruct with any model ID from the SambaCloud models page.

Example use case

To see ADK and SambaCloud working together, check out the multimodal agent example. This example demonstrates how to:

Process images

Understand and reason over visual inputs inside an ADK workflow.

Use external tools

Call APIs and services at any step of the agent’s reasoning loop.

Mockup to web page

Convert a visual design into a simple, functional web page automatically.

Troubleshooting

SAMBANOVA_API_KEY not found or authentication error

Confirm the variable is exported in your current shell:

echo $SAMBANOVA_API_KEY

Re-export if empty: export SAMBANOVA_API_KEY="your-api-key". Verify the key is active in the SambaCloud portal.

litellm cannot connect to SambaCloud

Confirm litellm is installed in your active virtual environment:

pip show litellm

If missing, run pip install litellm. Ensure your virtual environment is activated before running the agent.

ADK tool call fails or returns unexpected format

ADK expects tools to return serializable values. Check that your tool functions return strings, dicts, or lists — not custom objects. See the ADK tool documentation for supported return types.

Additional resources

ADK multimodal example

Runnable source code for the multimodal agent.

ADK documentation

Official reference for agents, tools, and advanced features.

Next steps: Swap the multimodal example’s default model for any SambaCloud-supported model and extend the agent with your own tools.

Overview

Developer tools

Frameworks

Interfaces and low-code

Data and infrastructure

Multimodal

Evaluation and monitoring

ADK integration guide

Multimodal input

Tool usage

Visual to code

Model-agnostic

Prerequisites

Setup

Configure the model

Example use case

Process images

Use external tools

Mockup to web page

Troubleshooting

Additional resources

ADK multimodal example

ADK documentation

Multimodal input

Tool usage

Visual to code

Model-agnostic

​Prerequisites

​Setup

​Configure the model

​Example use case

Process images

Use external tools

Mockup to web page

​Troubleshooting

​Additional resources

ADK multimodal example

ADK documentation

Prerequisites

Setup

Configure the model

Example use case

Troubleshooting

Additional resources