axonflow 6.0.0

AxonFlow Python SDK - Enterprise AI Governance in 3 Lines of Code

AxonFlow Python SDK

Enterprise AI Governance in 3 Lines of Code.

Evaluating AxonFlow in production? We're opening limited Design Partner slots.

Free 30-minute architecture and incident-readiness review, priority issue triage, roadmap input, and early feature access.

Apply here or email [e-mail].

No commitment required. We reply within 48 hours.

Questions or feedback?

Comment in GitHub Discussions or email [e-mail] for private feedback.

How This SDK Fits with AxonFlow

This SDK is a client library for interacting with a running AxonFlow control plane. It is used from application or agent code to send execution context, policies, and requests at runtime.

A deployed AxonFlow platform (self-hosted or cloud) is required for end-to-end AI governance. SDKs alone are not sufficient—the platform and SDKs are designed to be used together.

Architecture Overview (2 min)

If you're new to AxonFlow, this short video shows how the control plane and SDKs work together in a real production setup:

▶️ Watch on YouTube

Installation

pip install axonflow

With LLM provider support:

pip install axonflow[openai]      # OpenAI integration
pip install axonflow[anthropic]   # Anthropic integration
pip install axonflow[all]         # All integrations

Evaluation Tier (Free License)

Need more capacity than Community without moving to Enterprise? Evaluation uses the same core features with higher limits:

Limit	Community	Evaluation (Free)	Enterprise
Tenant policies	20	50	Unlimited
Org-wide policies	0	5	Unlimited
Audit retention	3 days	14 days	3650 days
Concurrent executions	5	25	Unlimited
Pending execution approvals	5	25	Unlimited

Concurrent executions applies to MAP and WCP executions per tenant. Pending execution approvals applies to MAP confirm/step mode and WCP approval queues.

Get a free Evaluation license · Full feature matrix

Quick Start

Async Usage (Recommended)

import asyncio
from axonflow import AxonFlow

async def main():
    async with AxonFlow(
        endpoint="https://your-agent.axonflow.com",
        client_id="your-client-id",
        client_secret="your-client-secret"
    ) as client:
        # Execute a governed query
        response = await client.proxy_llm_call(
            user_token="user-jwt-token",
            query="What is AI governance?",
            request_type="chat"
        )
        print(response.data)

asyncio.run(main())

Sync Usage

from axonflow import AxonFlow

with AxonFlow.sync(
    endpoint="https://your-agent.axonflow.com",
    client_id="your-client-id",
    client_secret="your-client-secret"
) as client:
    response = client.proxy_llm_call(
        user_token="user-jwt-token",
        query="What is AI governance?",
        request_type="chat"
    )
    print(response.data)

Features

Gateway Mode

For lowest-latency LLM calls with full governance and audit compliance:

from axonflow import AxonFlow, TokenUsage

async with AxonFlow(...) as client:
    # 1. Pre-check: Get policy approval
    ctx = await client.get_policy_approved_context(
        user_token="user-jwt",
        query="Find patient records",
        data_sources=["postgres"]
    )

    if not ctx.approved:
        raise Exception(f"Blocked: {ctx.block_reason}")

    # 2. Make LLM call directly (your code)
    llm_response = await openai.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": str(ctx.approved_data)}]
    )

    # 3. Audit the call
    await client.audit_llm_call(
        context_id=ctx.context_id,
        response_summary=llm_response.choices[0].message.content[:100],
        provider="openai",
        model="gpt-4",
        token_usage=TokenUsage(
            prompt_tokens=llm_response.usage.prompt_tokens,
            completion_tokens=llm_response.usage.completion_tokens,
            total_tokens=llm_response.usage.total_tokens
        ),
        latency_ms=250
    )

OpenAI Integration

Transparent governance for existing OpenAI code:

from openai import OpenAI
from axonflow import AxonFlow
from axonflow.interceptors.openai import wrap_openai_client

openai = OpenAI()
axonflow = AxonFlow(...)

# Wrap client - governance is now automatic
wrapped = wrap_openai_client(openai, axonflow, user_token="user-123")

# Use as normal
response = wrapped.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}]
)

MCP Connectors

Query data through MCP connectors:

# List available connectors
connectors = await client.list_connectors()

# Query a connector
result = await client.query_connector(
    user_token="user-jwt",
    connector_name="postgres",
    operation="query",
    params={"sql": "SELECT * FROM users LIMIT 10"}
)

MCP Policy Features (v3.2.0)

Exfiltration Detection - Prevent large-scale data extraction:

# Query with exfiltration limits (default: 10K rows, 10MB)
result = await client.query_connector(
    user_token="user-jwt",
    connector_name="postgres",
    operation="query",
    params={"sql": "SELECT * FROM customers"}
)

# Check exfiltration info
if result.policy_info.exfiltration_check.exceeded:
    print(f"Limit exceeded: {result.policy_info.exfiltration_check.limit_type}")

# Configure: MCP_MAX_ROWS_PER_QUERY=1000, MCP_MAX_BYTES_PER_QUERY=5242880

Dynamic Policy Evaluation - Orchestrator-based rate limiting, budget controls:

# Response includes dynamic policy info when enabled
if result.policy_info.dynamic_policy_info.orchestrator_reachable:
    print(f"Policies evaluated: {result.policy_info.dynamic_policy_info.policies_evaluated}")
    for policy in result.policy_info.dynamic_policy_info.matched_policies:
        print(f"  {policy.policy_name}: {policy.action}")

# Enable: MCP_DYNAMIC_POLICIES_ENABLED=true

Multi-Agent Planning

Generate and execute multi-agent plans:

# Generate a plan
plan = await client.generate_plan(
    query="Book a flight and hotel for my trip to Paris",
    domain="travel"
)

print(f"Plan has {len(plan.steps)} steps")

# Execute the plan
result = await client.execute_plan(plan.plan_id)
print(f"Result: {result.result}")

Configuration

from axonflow import AxonFlow, Mode, RetryConfig

client = AxonFlow(
    endpoint="https://your-agent.axonflow.com",
    client_id="your-client-id",               # Required for enterprise features
    client_secret="your-client-secret",       # Required for enterprise features
    mode=Mode.PRODUCTION,                     # or Mode.SANDBOX
    debug=True,                               # Enable debug logging
    timeout=60.0,                             # Request timeout in seconds
    retry_config=RetryConfig(                 # Retry configuration
        enabled=True,
        max_attempts=3,
        initial_delay=1.0,
        max_delay=30.0,
    ),
    cache_enabled=True,                       # Enable response caching
    cache_ttl=60.0,                           # Cache TTL in seconds
)

Error Handling

from axonflow.exceptions import (
    AxonFlowError,
    PolicyViolationError,
    AuthenticationError,
    RateLimitError,
    TimeoutError,
)

try:
    response = await client.proxy_llm_call(...)
except PolicyViolationError as e:
    print(f"Blocked by policy: {e.block_reason}")
except RateLimitError as e:
    print(f"Rate limited: {e.limit}/{e.remaining}, resets at {e.reset_at}")
except AuthenticationError:
    print("Invalid credentials")
except TimeoutError:
    print("Request timed out")
except AxonFlowError as e:
    print(f"AxonFlow error: {e.message}")

Response Types

All responses are Pydantic models with full type hints:

from axonflow import (
    ClientResponse,
    PolicyApprovalResult,
    PlanResponse,
    ConnectorResponse,
)

# Full autocomplete and type checking support
response: ClientResponse = await client.proxy_llm_call(...)
print(response.success)
print(response.data)
print(response.policy_info.policies_evaluated)

Development

# Install dev dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run linting
ruff check .
ruff format .

# Run type checking
mypy axonflow

Examples

Complete working examples for all features are available in the examples folder.

Community Features

# PII Detection - Automatically detect sensitive data
result = await client.get_policy_approved_context(
    user_token="user-123",
    query="My SSN is 123-45-6789"
)
# result.approved = True, result.requires_redaction = True (SSN detected)

# SQL Injection Detection - Block malicious queries
result = await client.get_policy_approved_context(
    user_token="user-123",
    query="SELECT * FROM users; DROP TABLE users;"
)
# result.approved = False, result.block_reason = "SQL injection detected"

# Static Policies - List and manage built-in policies
policies = await client.list_policies()
# Returns: [Policy(name="pii-detection", enabled=True), ...]

# Dynamic Policies - Create runtime policies
await client.create_dynamic_policy(
    name="block-competitor-queries",
    conditions={"contains": ["competitor", "pricing"]},
    action="block"
)

# MCP Connectors - Query external data sources
resp = await client.query_connector(
    user_token="user-123",
    connector_name="postgres-db",
    operation="query",
    params={"sql": "SELECT name FROM customers"}
)

# Multi-Agent Planning - Orchestrate complex workflows
plan = await client.generate_plan(
    query="Research AI governance regulations",
    domain="legal"
)
result = await client.execute_plan(plan.plan_id)

# Audit Logging - Track all LLM interactions
await client.audit_llm_call(
    context_id=ctx.context_id,
    response_summary="AI response summary",
    provider="openai",
    model="gpt-4",
    token_usage=TokenUsage(prompt_tokens=100, completion_tokens=200, total_tokens=300),
    latency_ms=450
)

Enterprise Features

These features require an AxonFlow Enterprise license:

# Code Governance - Automated PR reviews with AI
pr_result = await client.review_pull_request(
    repo_owner="your-org",
    repo_name="your-repo",
    pr_number=123,
    check_types=["security", "style", "performance"]
)

# Cost Controls - Budget management for LLM usage
budget = await client.get_budget("team-engineering")
# Returns: Budget(limit=1000.00, used=234.56, remaining=765.44)

# MCP Policy Enforcement - Automatic PII redaction in connector responses
resp = await client.query_connector("user", "postgres", "SELECT * FROM customers", {})
# resp.policy_info.redacted = True
# resp.policy_info.redacted_fields = ["ssn", "credit_card"]

For enterprise features, contact [e-mail].

Documentation

• Getting Started
• Gateway Mode Guide

Support

• Documentation: https://docs.getaxonflow.com
• Issues: https://github.com/getaxonflow/axonflow-sdk-python/issues
• Email: [e-mail]

If you are evaluating AxonFlow in a company setting and cannot open a public issue, you can share feedback or blockers confidentially here: Anonymous evaluation feedback form

No email required. Optional contact if you want a response.

Telemetry

This SDK sends anonymous usage telemetry (SDK version, OS, enabled features) to help improve AxonFlow. No prompts, payloads, or PII are ever collected. Opt out: AXONFLOW_TELEMETRY=off or DO_NOT_TRACK=1. See Telemetry Documentation for full details.

License

MIT - See LICENSE for details.