kakunin — Python SDK for AI agent identity & compliance

Python SDK for the Kakunin AI agent compliance API. Issues X.509 certificates to AI agents, monitors behavioral baselines, and enforces scope limits at the tool layer — with drop-in guards for LangChain, LlamaIndex, CrewAI, and AutoGen. Talks to the hosted service; free sandbox keys, no self-hosting required. Need your own control plane? See kakunin-core.

pip install kakunin

Requires Python 3.10+.

---

Quick Start

import asyncio
from kakunin import Kakunin

async def main():
    async with Kakunin(api_key="kak_live_...") as client:
        # Register an agent
        agent = await client.agents.create(
            name="TradeBot-1",
            model="gpt-4o",
            version="2024-11",
        )

        # Record a behavioral event — returns risk_band
        event = await client.events.create(
            agent_id=agent.id,
            action_type="transaction_initiated",
            details={"amount_usd": 50000, "venue": "NYSE"},
        )
        print(event.risk_band)  # "low" | "medium" | "high"

asyncio.run(main())

---

verify_agent_scope Decorator

Wrap any function (sync or async) to verify that the agent is active and holds required scopes before execution. Raises ScopeViolationError on failure — never swallows it.

from kakunin import Kakunin, verify_agent_scope

client = Kakunin(api_key="kak_live_...")

@verify_agent_scope(client, agent_id="agt-123", required_scopes=["trade.execute"])
async def execute_trade(order: dict) -> dict:
    # Only runs if agent agt-123 is active and has trade.execute scope
    return await broker.submit(order)

# Also works on sync functions
@verify_agent_scope(client, agent_id="agt-123", required_scopes=["data.read"])
def fetch_prices(symbol: str) -> list[float]:
    return market_data.get(symbol)

---

LangChain Integration

Option A — KakuninToolGuard (per-tool scope enforcement)

Wraps any LangChain tool. Scope is verified before every _run / _arun call.

from langchain_core.tools import tool
from kakunin import Kakunin
from kakunin.integrations.langchain import KakuninToolGuard

client = Kakunin(api_key="kak_live_...")

@tool
def execute_trade(order: str) -> str:
    """Execute a trade order on the exchange."""
    return broker.submit(order)

# Wrap the tool — drop into any LangChain agent as-is
guarded_trade = KakuninToolGuard(
    kakunin=client,
    agent_id="agt-123",
    tool=execute_trade,
    required_scopes=["trade.execute"],
)

from langchain.agents import create_react_agent
agent = create_react_agent(llm, tools=[guarded_trade], prompt=prompt)

Option B — verify_agent_scope on tool functions

from langchain_core.tools import tool
from kakunin import Kakunin, verify_agent_scope

client = Kakunin(api_key="kak_live_...")

@tool
@verify_agent_scope(client, agent_id="agt-123", required_scopes=["trade.execute"])
async def execute_trade(order: str) -> str:
    """Execute a trade order."""
    return await broker.submit(order)

Option C — Chain-level guard via callback

Guards an entire LangChain chain, not individual tools.

from kakunin.integrations.langchain import langchain_scope_callback

guard = langchain_scope_callback(
    client, agent_id="agt-123", required_scopes=["trade.execute"]
)
chain = my_chain.with_config(callbacks=[guard])

---

AutoGen Integration

KakuninConversableAgent subclasses AutoGen's ConversableAgent. It emits behavioral events for every message and verifies scope before each reply.

from autogen import UserProxyAgent
from kakunin import Kakunin
from kakunin.integrations.autogen import KakuninConversableAgent

client = Kakunin(api_key="kak_live_...")

risk_agent = KakuninConversableAgent(
    kakunin=client,
    agent_id="agt-456",
    required_scopes=["risk.assess"],   # verified before every generate_reply()
    # Standard AutoGen kwargs:
    name="RiskEngine",
    system_message="You are a risk analyst. Assess the given trade for regulatory risk.",
    llm_config={"model": "gpt-4o"},
)

user_proxy = UserProxyAgent(name="User", human_input_mode="NEVER")
user_proxy.initiate_chat(risk_agent, message="Assess trade T-984231")

To attach the agent's certificate serial to outbound HTTP calls:

from kakunin.integrations.autogen import KakuninConversableAgent, KakuninHttpxMixin

class MyAgent(KakuninHttpxMixin, KakuninConversableAgent):
    pass

agent = MyAgent(
    kakunin=client,
    agent_id="agt-456",
    cert_serial="3A:F2:...",
    name="RiskEngine",
    llm_config={"model": "gpt-4o"},
)

# Outbound requests carry X-Kakunin-Cert-Serial automatically
resp = await agent.http_client.get("https://internal-service/prices")

---

Other Framework Integrations

Framework	Import	What it does
LangGraph	kakunin.integrations.langgraph.kakunin_node	Decorator: emits behavioral event on every node execution
LlamaIndex	kakunin.integrations.llamaindex.KakuninFunctionToolGuard	Wraps FunctionTool with scope check
CrewAI	kakunin.integrations.crewai.KakuninCrewAgent	Agent subclass with event emission + scope guard
CAMEL-AI	kakunin.integrations.camel.KakuninToolkit	4 FunctionTools (verify status, check scope, risk score, emit event)

---

Error Handling

from kakunin import ScopeViolationError, RateLimitError, AuthenticationError

try:
    result = await guarded_function()
except ScopeViolationError as e:
    # Agent is suspended, retired, or missing a required scope
    print(f"Scope check failed: {e.missing_scopes}")
    # Do NOT proceed — the agent is not authorised
except RateLimitError:
    await asyncio.sleep(1)
    # Retry
except AuthenticationError:
    # Invalid API key — check KAKUNIN_API_KEY env var
    raise

---

Full docs at kakunin.ai/docs.

Contributors

Thanks to everyone who contributes (emoji key) — code and non-code alike:

Contributions of any kind are welcome — this project follows the all-contributors spec.