Use Atomic Agents with Eden AI

Atomic Agents sends its structured-output requests through an Instructor-wrapped OpenAI client, so pointing that client at https://api.edenai.run/v3 is the whole integration.

Overview

Atomic Agents is a lightweight, modular Python framework for building AI agents on composable, schema-driven building blocks (via Instructor and Pydantic). It works with Eden AI through its existing OpenAI-compatible client, so your agents reach models from OpenAI, Anthropic, Google, Mistral and more behind one key.

Installation

Eden AI reuses the OpenAI client and Instructor that Atomic Agents already relies on. Atomic Agents 2 needs Python 3.12 or later; on an older Python, pip installs the 1.x line, which has no AtomicAgent or AgentConfig.

pip install "atomic-agents>=2" openai instructor

poetry add "atomic-agents>=2" openai instructor

Quick Start

Atomic Agents does not read an API key from the environment itself, so the snippets below read EDEN_AI_API_KEY. Export it first:

export EDEN_AI_API_KEY="YOUR_API_KEY"

Wrap an OpenAI client pointed at Eden AI with Instructor, the same pattern used for any other OpenAI-compatible provider, and pass it to your agent:

import os
import instructor
from openai import OpenAI
from atomic_agents import (
    AgentConfig,
    AtomicAgent,
    BasicChatInputSchema,
    BasicChatOutputSchema,
)

client = instructor.from_openai(
    OpenAI(
        base_url="https://api.edenai.run/v3",
        api_key=os.getenv("EDEN_AI_API_KEY"),
    )
)

agent = AtomicAgent[BasicChatInputSchema, BasicChatOutputSchema](
    config=AgentConfig(
        client=client,
        model="vertex/gemini-flash-latest",  # <provider>/<model>
    )
)

response = agent.run(BasicChatInputSchema(chat_message="Hello! How are you?"))
print(response.chat_message)

Atomic Agents asks Instructor for structured output. Its default TOOLS mode sends a tool call, so pick models that support function calling. Every model named on this page does.

Model naming

Use the provider/model format, for example vertex/gemini-flash-latest, anthropic/claude-sonnet-latest or mistral/mistral-large-latest. To browse every id, see List LLM Models.

Switching models

Change the model string on AgentConfig to reach any Eden AI model, with no per-provider SDKs and no code changes:

import os
import instructor
from openai import OpenAI
from atomic_agents import (
    AgentConfig,
    AtomicAgent,
    BasicChatInputSchema,
    BasicChatOutputSchema,
)

client = instructor.from_openai(
    OpenAI(
        base_url="https://api.edenai.run/v3",
        api_key=os.getenv("EDEN_AI_API_KEY"),
    )
)

for model in [
    "openai/gpt-latest",
    "anthropic/claude-sonnet-latest",
    "mistral/mistral-large-latest",
]:
    agent = AtomicAgent[BasicChatInputSchema, BasicChatOutputSchema](
        config=AgentConfig(client=client, model=model)
    )
    response = agent.run(BasicChatInputSchema(chat_message="Write a haiku about the sea."))
    print(model, "->", response.chat_message)

Environment Variables

Atomic Agents reads no Eden AI variable of its own. EDEN_AI_API_KEY is the name the snippets on this page use, so you can rename it as long as you change the os.getenv call to match.

Next Steps

aisuite - Another OpenAI-style multi-provider interface
Chat Completions - Core LLM endpoint
List LLM Models - Browse available providers and models

V3 Documentation

Quick Start

Overview

LLMs

Expert Models

General

Organization

Data Governance

Integrations

Overview

Installation

Quick Start

Model naming

Switching models

Environment Variables

Next Steps

​Overview

​Installation

​Quick Start

​Model naming

​Switching models

​Environment Variables

​Next Steps

Overview

Installation

Quick Start

Model naming

Switching models

Environment Variables

Next Steps