Drop-in tool classes for the two most common Python agent frameworks. About 80 lines of code each, no install required beyond the framework itself. If you've already built a LangChain or LlamaIndex agent, you can plug AegisAI in this afternoon.
BaseTool subclass — drop into any agent.pip install langchain langchain-openai httpx
export OPENAI_API_KEY=sk-...
export AEGISAI_USER_TOKEN=<JWT>Copy aegisai_tool.py into your project. Use it like any other LangChain tool — ReAct, OpenAI Functions, structured-tool, LangGraph, all work.
from aegisai_tool import AegisAITool
tool = AegisAITool(
base_url="https://api.aegisai.store",
user_token=os.environ["AEGISAI_USER_TOKEN"],
tenant_id="acme",
)
agent = create_react_agent(llm, [tool], prompt)FunctionTool factory — same shape, different framework.pip install llama-index llama-index-llms-openai httpx
export OPENAI_API_KEY=sk-...
export AEGISAI_USER_TOKEN=<JWT>Copy aegisai_tool.py from samples/connectors/llamaindex/. Wraps the same POST /query endpoint behind a LlamaIndex FunctionTool.
from aegisai_tool import make_aegisai_tool
tool = make_aegisai_tool(
base_url="https://api.aegisai.store",
user_token=os.environ["AEGISAI_USER_TOKEN"],
tenant_id="acme",
)
agent = ReActAgent.from_tools([tool], llm=llm)The tool class holds the user's JWT. If you build a web service where one process handles requests for multiple humans, instantiate the tool fresh per request — never cache it across users:
# inside your request handler
tool = AegisAITool(
base_url=AEGIS_BASE,
user_token=current_user.jwt,
tenant_id=current_user.tenant_id,
)
agent = create_react_agent(llm, [tool], prompt)
result = agent.invoke({"input": user_message})If you cache and reuse one tool across users, you'll cross-attribute the audit log — Alice's tool will report Bob's queries. AegisAI's row-level tenant policy will still block cross-tenant reads, but the audit chain becomes useless for incident response.