Skip to content
Navigation

Migrating from AWorld to Orbiter

This guide maps AWorld concepts and imports to their Orbiter equivalents.

This guide maps AWorld concepts and imports to their Orbiter equivalents.

Package Mapping

AWorldOrbiterNotes
aworld (monolith)orbiter (meta-package)Split into 13 focused packages
aworld.agentsorbiter.agentSingle Agent class replaces LLMAgent, TaskLLMAgent, etc.
aworld.core.toolorbiter.tool@tool decorator, FunctionTool, Tool ABC
aworld.runnerorbiter.runnerrun(), run.sync(), run.stream()
aworld.modelsorbiter.modelsget_provider("openai:gpt-4o") factory
aworld.core.context.amniorbiter.contextClean rewrite — neurons, processors, workspace
aworld.memoryorbiter.memoryShort/long-term, SQLite/Postgres backends
aworld.mcp_clientorbiter.mcpMCP client + @mcp_server decorator
aworld.sandboxorbiter.sandboxLocal + Kubernetes sandboxes
aworld.traceorbiter.traceOpenTelemetry-based tracing
aworld.evaluationsorbiter.evalScorers, reflection, evaluator
aworld.ralph_looporbiter.ralphRalph loop — state, detectors, runner
aworld.experimental.a2aorbiter.a2aAgent-to-Agent protocol

Agent Definition

AWorld:

python
from aworld.agents import LLMAgent
from aworld.config.conf import AgentConfig

config = AgentConfig(name="my-agent", llm_provider="openai", llm_model_id="gpt-4o")
agent = LLMAgent(agent_config=config, task_config=task_config)

Orbiter:

python
from orbiter import Agent

agent = Agent(name="my-agent", model="openai:gpt-4o", instructions="...", tools=[...])

Running Agents

AWorld:

python
from aworld.runner import create_runner
runner = create_runner(agent_config=config, task_config=task_config)
result = await runner.run(task)

Orbiter:

python
from orbiter import run

result = await run(agent, "What is 2+2?")      # async
result = run.sync(agent, "What is 2+2?")        # sync
async for event in run.stream(agent, "prompt"):  # streaming
    print(event)

Tools

AWorld:

python
from aworld.tools.function_tools import FunctionTool
tool = FunctionTool(name="search", func=search_fn, description="...")

Orbiter:

python
from orbiter import tool

@tool
async def search(query: str) -> str:
    """Search the web."""
    return "results"

Multi-Agent (Swarm)

AWorld:

python
from aworld.agents import SwarmComposerAgent
swarm = SwarmComposerAgent(agents=[a, b], swarm_config=config)

Orbiter:

python
from orbiter import Swarm
swarm = Swarm(agents=[a, b], mode="workflow")  # or "handoff", "team"

Context

AWorld:

python
from aworld.core.context.amni.contexts import AmniContext
from aworld.core.context.amni.config import AmniConfig

Orbiter:

python
from orbiter.context import Context, ContextConfig
ctx = Context(config=ContextConfig(automation_level="copilot"))

Key Differences

  1. Single Agent class — AWorld had LLMAgent, TaskLLMAgent, LoopLLMAgent, ParallelLLMAgent, SerialLLMAgent. Orbiter has one Agent with composable behavior via Swarm modes and agent groups.

  2. Model strings — use "provider:model" format (e.g., "openai:gpt-4o", "anthropic:claude-sonnet-4-20250514") instead of separate provider/model config fields.

  3. Async-first — all agent execution is async. Use run.sync() for synchronous contexts.

  4. Modular packages — install only what you need. orbiter-core has zero heavy dependencies.

  5. Type safety — full pyright strict-mode compliance across all packages.