AIM Documentation

AIM (Agent Identity Management) gives every AI agent a cryptographic identity, capability-based authorization, and a tamper-evident audit trail. It is for developers and platform teams who need to prove which agent took an action, confirm it was authorized, and keep a verifiable record for compliance. Install the SDK with pip install aim-sdk, or run identity and trust commands through the OpenA2A CLI with npm install -g opena2a-cli.

pip install aim-sdk
aim-sdk login

1. Choose your setup

docker compose up -d

npm install @opena2a/aim-core

2. Pick your path

3. Key concepts

5-step Fine-Grained Authorization

On the server, every privileged action runs through five checks before execution. The first four carry a low latency budget. Step 5 calls the NanoMind security classifier, a 3M-parameter local Mamba model, with a larger budget reserved for high-risk operations. No external calls.

Risk detection in the Python SDK

@perform_action auto-detects the risk level from the capability string using two lookup tables. A namespace prefix sets a baseline, an action suffix adjusts it, and when the two disagree the higher risk wins. A SPECIFIC_CAPABILITY_MAP overrides both for known patterns (for example user:delete escalates to critical). Pass risk_level="critical" to override, and jit_access=True to pause execution until a human approves in the dashboard.

Framework and LLM auto-detection

secure() auto-detects the framework from imports (langchain, crewai, llama_index, anthropic, openai). When both a framework and an LLM provider are present, the framework wins.

# python
from aim_sdk import secure

# langchain import present alongside an openai import:
# the framework (langchain) wins over the LLM provider (openai)
agent = secure("my-first-agent")

Server integrations

Self-hosted servers can forward audit log entries to Splunk HEC and Microsoft Sentinel. This forwarder is not available in AIM Cloud. A CyberArk connector is planned.

Two trust-scoring models

The local and server trust scores measure different things, and both can run for the same agent when local-to-server reporting is enabled. The local 8-factor score (TypeScript) answers whether the agent's security posture is set up correctly, computed from local files. The server 9-factor score (Go) answers whether the agent is behaving in a way that should still be trusted right now, updated on every action.