yantrix-adapters 1.0.0

Cross-framework adapters for Yantrix agent infrastructure — LangChain, CrewAI, AutoGen, MCP

yantrix-adapters

Cross-framework adapters for Yantrix agent infrastructure.

Add policy enforcement, semantic memory, and full audit traces to any agent framework in 5 lines.

pip install yantrix-adapters

---

Supported Frameworks

Framework	Class	Install
LangChain	YantrixToolkit	pip install yantrix-adapters[langchain]
CrewAI	YantrixCrew	pip install yantrix-adapters[crewai]
AutoGen	YantrixAssistantAgent	pip install yantrix-adapters[autogen]
MCP	YantrixMCPRouter	pip install yantrix-adapters[mcp]
All	—	pip install yantrix-adapters[all]

---

LangChain

from langchain.tools import DuckDuckGoSearchRun, WikipediaQueryRun
from yantrix_adapters.langchain import YantrixToolkit

# Before: tools = [DuckDuckGoSearchRun(), WikipediaQueryRun()]
# After:
ytx = YantrixToolkit(api_key="ytx_...", agent_id="research_agent")
tools = ytx.wrap([DuckDuckGoSearchRun(), WikipediaQueryRun()])

# Drop into your existing agent — nothing else changes
agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION)

Observability-only (no tool replacement):

from yantrix_adapters.langchain import YantrixCallbackHandler

handler = YantrixCallbackHandler(api_key="ytx_...")
agent = initialize_agent(tools, llm, callbacks=[handler])

---

CrewAI

from crewai import Crew, Agent, Task
from yantrix_adapters.crewai import YantrixCrew

researcher = Agent(role="researcher", goal="find information", ...)
writer = Agent(role="writer", goal="write content", ...)
task = Task(description="Research and write about AI trends", ...)

# Before: crew = Crew(agents=[researcher, writer], tasks=[task])
# After:
crew = YantrixCrew.wrap(
    Crew(agents=[researcher, writer], tasks=[task]),
    api_key="ytx_...",
    agent_id="content_crew",
)

result = crew.kickoff()  # Identical call — now governed

---

AutoGen

from yantrix_adapters.autogen import YantrixAssistantAgent, YantrixUserProxy

# Drop-in replacements for AssistantAgent and UserProxyAgent
assistant = YantrixAssistantAgent(
    "assistant",
    ytx_api_key="ytx_...",
    llm_config={"config_list": config_list},
)

user_proxy = YantrixUserProxy(
    "user_proxy",
    ytx_api_key="ytx_...",
    human_input_mode="NEVER",
    code_execution_config={"work_dir": "coding"},
)

user_proxy.initiate_chat(assistant, message="Research OpenAI pricing trends")

For GroupChat:

from autogen import GroupChat, GroupChatManager
from yantrix_adapters.autogen import YantrixGroupChat

chat = YantrixGroupChat.wrap(
    GroupChat(agents=[agent1, agent2], messages=[], max_round=10),
    api_key="ytx_...",
)
manager = GroupChatManager(groupchat=chat, llm_config=llm_config)

---

MCP

from mcp.server import Server
from yantrix_adapters.mcp import YantrixMCPRouter

server = Server("my-tools")
router = YantrixMCPRouter(api_key="ytx_...", agent_id="mcp_server")

@server.call_tool()
async def call_tool(name: str, arguments: dict):
    # Route through Yantrix — policy check, memory, trace
    return await router.route(name, arguments, handler=_original_handler)

async def _original_handler(name, arguments):
    # Your existing tool logic here
    ...

FastAPI middleware (wraps entire MCP HTTP server):

from fastapi import FastAPI
from yantrix_adapters.mcp import YantrixMCPMiddleware

app = FastAPI()
app.add_middleware(YantrixMCPMiddleware, api_key="ytx_...", agent_id="mcp_server")

---

What happens on every call

Your agent action
      ↓
Yantrix Policy Engine  ← is this allowed? within budget?
      ↓ (if allowed)
Yantrix Memory         ← inject relevant past context
      ↓
Original handler       ← your tool/task/agent runs normally
      ↓
Yantrix Trace          ← full audit log stored
      ↓
Result returned

All traces visible at trace.yantrix.ai.

---

Configuration

ytx = YantrixToolkit(
    api_key="ytx_...",           # or YANTRIX_API_KEY env var
    agent_id="my_agent",         # or YANTRIX_AGENT_ID env var
    budget_usd=5.0,              # optional daily budget cap
    timeout=30,                  # request timeout seconds
    fallback_on_error=True,      # if True: proceed if Yantrix is unreachable
)

fallback_on_error=True (default) means your agents keep working even if Yantrix has an outage. Set to False for strict enforcement.

---

Links

• Yantrix Runtime API
• Trace Dashboard
• Full Docs
• GitHub