Multi-Framework Support

Serve agents built with the Claude Agent SDK, LangGraph, DSPy and Antigravity from one AgentOS.

v2.6.0

AgentOS runs agents built with Agno, Claude Agent SDK, LangGraph, DSPy and Antigravity through one runtime, one API and one UI. Each adapter wraps an external agent so it satisfies Agno's AgentProtocol and behaves like a native Agent for routing, streaming and session persistence.

Install the combined example's dependencies and set both provider keys in the server terminal:

uv pip install -U "agno[os]" openai claude-agent-sdk
export OPENAI_API_KEY="your_openai_api_key"
export ANTHROPIC_API_KEY="your_anthropic_api_key"

Save the example as multi_framework.py and run python multi_framework.py.

from agno.agent import Agent
from agno.agents.claude import ClaudeAgent
from agno.db.sqlite import SqliteDb
from agno.os import AgentOS
from agno.tools.workspace import Workspace

claude_agent = ClaudeAgent(
    name="Claude Code Agent",
    model="claude-sonnet-4-6",
    allowed_tools=["Read", "Edit", "Bash"],
    permission_mode="acceptEdits",
    max_turns=10,
)

agno_agent = Agent(
    name="Agno Agent",
    model="openai:gpt-5.4",
    tools=[Workspace(root=".", allowed=["read", "list", "search"])],
)

agent_os = AgentOS(
    agents=[agno_agent, claude_agent],
    tracing=True,
    db=SqliteDb(db_file="tmp/agentos.db"),
)
app = agent_os.get_app()

if __name__ == "__main__":
    agent_os.serve(app="multi_framework:app", reload=True)

Supported Frameworks

FrameworkAdapterInstall
Claude Agent SDKClaudeAgentuv pip install "agno[os]" claude-agent-sdk
LangGraphLangGraphAgentuv pip install "agno[os]" langgraph langchain-openai
DSPyDSPyAgentuv pip install "agno[os]" dspy
AntigravityAntigravityAgentuv pip install "agno[os]" (set GEMINI_API_KEY)

Once registered, an external agent is routed, streamed over SSE and persisted just like a native Agno agent. The same adapter also works standalone via .run() and .print_response(). Persistence of Agno records does not guarantee that an external framework resumes its conversation after a restart. For example, the Claude adapter keeps its SDK resume-session mapping in process memory. See the per-framework pages for the applicable behavior.

What Works and What Doesn't

The adapters cover the basics every AgentOS deployment needs (registration, streaming, sessions, tool visibility). Agno-specific capabilities like delegation, knowledge, dependencies, and hooks live on the native Agent and Team and are not available through external frameworks.

CapabilitySupportedNotes
AgentOS(agents=[...]) registration✅Adapters satisfy AgentProtocol
/agents and /agents/{id}/runs endpoints✅Same routes as native agents
SSE streaming✅Content and tool events depend on the adapter and framework
Session persistence✅Agno run/session records persist with db; framework resume behavior varies
Standalone .run() / .print_response()✅Sync and async
Tool call visibility in the UI✅Wrapped as Agno tool events
Use as a Team member-Agno Team orchestration is built around the native Agent
Memory, knowledge, dependencies, hooks, guardrails-These are Agno-SDK concepts wired into the native Agent
Structured input and output-Use the framework's own typing (DSPy signatures, LangGraph state)
Skills, reasoning, learning-Native Agent only

Next Steps

Claude Agent SDK

Run Claude Code as an AgentOS agent.

LangGraph

Wrap a compiled LangGraph graph.

DSPy

Serve a DSPy program (Predict, ChainOfThought, ReAct).

Antigravity

Run Google's Managed Agents (Gemini API) in a sandbox.

Developer Resources