Skip to main content
Atlas SDK follows a simple promise: your agent, our orchestration. Adapters create a small, consistent interface so you can place the Student-Teacher loop on top of nearly anything—from hosted APIs to local Python functions.

​
Choosing an Adapter

AdapterUse when…StrengthsThings to watch
openaiYou have an OpenAI or Azure OpenAI chat model.Minimal setup, native tool calling, streaming support.Only works with OpenAI-compatible APIs.
http_apiYour agent already runs behind an HTTP endpoint.Language-agnostic, deploy-anywhere.You define the payload schema, handle auth, and parse responses.
pythonYou want to call local functions or LangChain runnables directly.Lowest latency, easy debugging.Runs inside the orchestrator process—ensure your code is safe and performant.

​
OpenAI Adapter (atlas/agent/openai_adapter.py)

This is the fastest way to get started, especially with GPT-4o-mini or Azure OpenAI models. 
agent:
  type: openai
  name: sdk-quickstart-openai
  system_prompt: |
    You are the Atlas Student. Be concise and helpful.
  tools: []
  llm:
    provider: openai
    model: gpt-4o-mini
    api_key_env: OPENAI_API_KEY
    temperature: 0.0
    max_output_tokens: 768
  • Supports conversation history and tool call metadata automatically.
  • Accepts response_format for JSON mode.
  • For Claude, Gemini, or other models, use the http_api adapter.

​
HTTP Adapter (atlas/agent/http_adapter.py)

Best for microservices, serverless functions, or non-Python agents. 
agent:
  type: http_api
  name: example-http-agent
  transport:
    base_url: https://agent.example.com/run
    headers:
      Authorization: Bearer $AGENT_TOKEN
    timeout_seconds: 60
    retry:
      attempts: 3
      backoff_seconds: 1.0
  payload_template:
    mode: inference
  result_path:
    - data
    - output
Wrap any service that accepts a JSON request. You define:
  • payload_template: The base JSON payload; the adapter injects the prompt and optional metadata.
  • result_path: A list of keys to traverse in the JSON response to find the agent’s output string.
  • Auth & retries: Handled via the transport block, which mirrors httpx settings.

​
Python Adapter (atlas/agent/python_adapter.py)

Ideal when your agent is a local Python function or a LangChain runnable. 
agent:
  type: python
  name: example-python-agent
  import_path: agents.python_example
  attribute: run_agent
  working_directory: ./
  allow_generator: false
  • The adapter imports the attribute (e.g., run_agent) from the import_path module.
  • If the callable is async, the adapter awaits it; sync functions run in a background thread.
  • Set allow_generator: true to let the adapter consume a generator, concatenating all yielded strings into a single output.

​
Building Custom Adapters

All adapters share a minimal interface (AgentAdapter). To add a new one (e.g., for gRPC), follow these steps:
  1. Extend the AdapterType enum in atlas/config/models.py.
  2. Implement a class inheriting from AgentAdapter.
  3. Register it in atlas/agent/__init__.py.
from atlas.agent.registry import AgentAdapter, register_adapter
from atlas.config.models import AdapterType

class GRPCAdapter(AgentAdapter):
    async def ainvoke(self, prompt: str, metadata: dict | None = None) -> str:
        # 1. Connect to your gRPC service.
        # 2. Build the request from the prompt.
        # 3. Execute the call and get a response.
        # 4. Return the response as a string.
        return f"Response for prompt: {prompt}"

# Assumes you've added GRPC to the AdapterType enum
register_adapter(AdapterType.GRPC, GRPCAdapter)
Most teams start by copying the http_api adapter and swapping the transport layer.

​
Decision Checklist

NeedRecommendation
Fastest time-to-first-runopenai adapter with sdk_quickstart.yaml.
Reuse an existing microservicehttp_api adapter with proper retries and auth.
Full control in local experimentspython adapter calling your local function.
Access non-OpenAI APIs (Claude, Gemini, etc.)Wrap them with the http_api adapter.

​
Next Steps

⌘I