MACAW Documentation
MACAW is a trust layer for AI systems. It provides cryptographic identity, policy enforcement, and audit logging implemented as a distributed mesh for agents, tools, and LLMs.
Architecture
MACAW creates a distributed trust mesh where endpoints register, discover each other, and communicate with cryptographic guarantees. The control plane provides identity, registry, policy, and logging services.
Every participant—whether an AI agent, tool, or LLM—becomes a verified endpoint that can independently enforce policies without trusting the caller.
┌─────────────────────────────────────────────────┐
│ YOUR APPLICATION │
├─────────────────────────────────────────────────┤
│ SecureOpenAI │ SecureAnthropic │ SecureMCP│
├─────────────────────────────────────────────────┤
│ MACAWClient │
└─────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────┐
│ TRUST LAYER CONTROL PLANE │
│ ┌────────┐ ┌────────┐ ┌────────┐ ┌────────┐ │
│ │Identity│ │Registry│ │ Policy │ │ Audit │ │
│ └────────┘ └────────┘ └────────┘ └────────┘ │
└─────────────────────────────────────────────────┘Security Model
MACAW implements three complementary mechanisms that work together to secure AI operations:
Authenticated Workflows
Every invocation is cryptographically signed by the caller and independently verified by the receiver. Policy enforcement happens at endpoints, not a central chokepoint.
Learn more →Authenticated Prompts
Prompts are signed and tracked through derivation chains. Permissions can only narrow as prompts evolve, providing defense against prompt injection attacks.
Learn more →Authenticated Context
Application state is protected with session-bound signatures. Enables multi-user isolation and safe delegation of authority in agentic workflows.
Learn more →Integration
MACAW provides drop-in adapters for common AI frameworks. Change your import, and your existing code gains cryptographic signing, policy enforcement, and audit logging.
# Before
from openai import OpenAI
client = OpenAI()
# After
from macaw_adapters.openai import SecureOpenAI
client = SecureOpenAI(app_name="my-app")
# Same API, now with policy enforcement
response = client.chat.completions.create(
model="gpt-4",
messages=[{"role": "user", "content": "Hello"}]
)Getting Started
Concepts
Authenticated Workflows
Signed invocations and independent verification
Authenticated Prompts
Lineage tracking and monotonic narrowing
Authenticated Context
Session isolation and state protection
Identity Bridge
Enterprise IDP integration
MAPL
Policy language for access control
Observability
Metrics, traces, and audit logs
SDK Reference
Framework Guides
Console
The MACAW Console is your administrative interface for managing AI security at scale. Monitor activity, manage policies, review audit logs, and configure workspace settings.