AutoGen

Install

pip install comply54[autogen]
# Installs autogen-agentchat ≥ 0.4 automatically.

Automatic enforcement (recommended)

Wrap tools before passing them to AssistantAgent. Every tool call is evaluated against the compliance pack before execution — blocked calls return a structured JSON error to the model without running the function.

from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
from comply54.autogen import comply54_tools
from comply54.sectors import NigeriaFintechCompliance

compliance = NigeriaFintechCompliance()
client = OpenAIChatCompletionClient(model="gpt-4o")

async def transfer_funds(recipient: str, amount: float, currency: str) -> str:
    """Transfer funds between accounts."""
    ...

async def check_balance(account: str) -> str:
    """Return account balance."""
    ...

agent = AssistantAgent(
    name="finance_agent",
    model_client=client,
    tools=comply54_tools([transfer_funds, check_balance], compliance),
)

When a tool call is blocked the model receives:

{
  "blocked": true,
  "decision": "deny",
  "reason": "CBN NIP: single-transfer limit exceeded",
  "regulation": "CBN NIP Framework 2023",
  "rule_triggered": "cbn_nip_single_transfer_limit",
  "citations": [{ "document": "CBN NIP Framework 2023", "section": "s.4.2", ... }],
  "audit_id": "aud_f1e2d3",
  "jurisdictions": ["NG"]
}

Wrap a single tool

from comply54.autogen import comply54_tool

safe_transfer = comply54_tool(
    transfer_funds,
    compliance,
    context={"kyc_tier": 3},       # optional — forwarded to every evaluation
    block_on_escalate=True,        # optional — treat escalate as deny
)

agent = AssistantAgent(
    name="finance_agent",
    model_client=client,
    tools=[safe_transfer, check_balance],
)

Self-check tool (passive)

Add a compliance_tool to the agent's tool list. The model decides when to call it — useful when you want the LLM to reason about compliance before acting rather than having the runtime block automatically.

from comply54.autogen import comply54_tool, compliance_tool

agent = AssistantAgent(
    name="finance_agent",
    model_client=client,
    tools=[
        comply54_tool(transfer_funds, compliance),
        compliance_tool(compliance),       # agent can self-check any action
    ],
)

The self-check tool returns:

{
  "overall": "escalate",
  "blocked": true,
  "audit_id": "aud_a1b2c3",
  "violations": [{ "pack": "cbn", "regulation": "CBN NIP Framework 2023", ... }]
}

Sector packs

from comply54.sectors import (
    NigeriaFintechCompliance,
    NigeriaHealthcareCompliance,
    KenyaFintechCompliance,
    PanAfricanFintechCompliance,
)

API reference

comply54_tool(fn, compliance, *, description, context, block_on_escalate, name)

Wrap a single callable with pre-execution compliance enforcement.

Parameter	Type	Default	Description
fn	Callable	—	The function to wrap (sync or async).
compliance	SectorCompliance	—	Compliance pack to evaluate against.
description	str	fn.__doc__	Tool description forwarded to FunctionTool.
context	dict	{}	Session context for every evaluation (e.g. {"kyc_tier": 3}).
block_on_escalate	bool	False	Treat "escalate" decisions as blocks.
name	str	fn.__name__	Override the tool name.

Returns a FunctionTool ready for AssistantAgent(tools=[...]).

---

comply54_tools(tools, compliance, *, context, block_on_escalate)

Bulk-wrap a list of callables or existing FunctionTool instances. All tools share the same compliance pack and context.

---

compliance_tool(compliance)

Create a passive self-check FunctionTool. The agent must call it explicitly.

---

Migrating from pyautogen ≤ 0.2

The UserProxyAgent.execute_function() override pattern is not supported in autogen-agentchat ≥ 0.4. Replace the old pattern with comply54_tools():

# Before (pyautogen ≤ 0.2)
from comply54.autogen import Comply54UserProxy
proxy = Comply54UserProxy(compliance, name="proxy", human_input_mode="NEVER")

# After (autogen-agentchat ≥ 0.4)
from comply54.autogen import comply54_tools
from autogen_agentchat.agents import AssistantAgent

agent = AssistantAgent(
    name="agent",
    model_client=client,
    tools=comply54_tools([your_tool], compliance),
)