Every AI framework
lets agents act.

None verify what they're allowed to do.

AIP is the identity and intent verification protocol for autonomous AI agents. Cryptographic passports. Scoped permissions. Real-time revocation. Two lines to integrate.

pip install aip-protocol
from aip_protocol import shield

@shield(actions=["read_data", "send_email"], limit=100.0)
class SalesAgent:
    def run(self):
        # Agent auto-gets Ed25519 passport + DID
        # Every action verified against boundaries
        # Kill switch available in real-time
        ...
Ed25519
Cryptography
8-Step
Verification Pipeline
<1ms
Revocation Latency
22
Structured Error Codes

Works with your stack.

First-class integrations for the frameworks you already use. One line to protect any agent.

LangChainReady

Wrap any LangChain tool with @aip_tool — automatic signing, boundary checks, and verification on every call.

@aip_tool(limit=500)
def transfer_funds(amount, to): ...
pip install aip-langchain
CrewAIReady

Protect CrewAI agents and tasks. AIP verification runs before any task execution.

researcher = aip_agent(agent,
  actions=["research"], limit=100)
pip install aip-crewai
AutoGenReady

Secure multi-agent conversations. Every message between AutoGen agents is AIP-verified.

assistant = aip_wrap(assistant,
  actions=["analyze"], limit=500)
pip install aip-autogen

The missing security layer.

AI agents are making real decisions with real money. The frameworks that power them have zero verification built in.

Without AIP
Agent drains a bank account — no spending limits enforced
Rogue agent emails 10,000 customers — no action boundary
Agent impersonates another — no cryptographic identity
Compromised agent runs for hours — no kill switch
With AIP
Monetary limits per transaction — $50 max, enforced cryptographically
Action allowlists — only approved operations execute
Ed25519 passport per agent — unforgeable identity
Kill switch revokes any agent instantly — zero propagation delay

Everything your agents need
to be trustworthy.

Six capabilities. One protocol. Works with any framework.

Identity
ED25519 + DID

Cryptographic Identity
& Trust Scoring

Every agent gets an unforgeable Ed25519 keypair and DID address. Bayesian trust scores built from verification history.

Safety
<1MS

Emergency Kill Switch

Revoke or suspend any agent globally. Propagates via real-time mesh network. Zero delay. No agent can act after revocation.

The Cage

Boundary Enforcement

Action allowlists, deny lists, monetary limits, geo-fencing.

API Keys

Scoped Permissions

9 granular scopes. Plan-based defaults from Free to Enterprise.

Verification

Tiered Verification

Tier 0 (<1ms) low-risk. Tier 1 (~5ms) attestation. Tier 2 (~100ms) cross-org. 8 checks per intent.

Built for developers who ship.

From first install to production in under 5 minutes.

4 commands

CLI-First Workflow

aip init scaffolds your agent. aip login authenticates. aip watch streams live verifications. All from your terminal.

@shield decorator

One-Line Protection

@shield auto-detects your class name, generates passports, and enforces boundaries. Zero boilerplate.

X-API-Key

Scoped API Keys

9 granular scopes. Use X-API-Key header for server-to-server. No JWT dance needed for automation.

Live UI

Real-Time Dashboard

See every verification, trust score change, and revocation as it happens. Full audit trail included.

Built-in

Analytics & Webhooks

Verification volume, approval rates, 30-day trends. Webhooks for verification.passed, revocation.issued, and more.

PyPI v0.3.0

Open Source SDK

MIT licensed. 63 tests. Zero dependencies beyond Python stdlib + cryptography. pip install and go.

Three steps to verified agents.

From zero to production-grade agent security.

01

Create Passport

Generate an Ed25519 identity with action boundaries, monetary limits, and a DID address. One decorator or one API call.

02

Sign Intents

Every action your agent takes is packaged into a cryptographically signed intent envelope with proof of authorization.

03

Verify & Execute

The 8-step verification pipeline checks identity, expiration, boundaries, trust score, and revocation status. Pass → execute. Fail → block.

Granular API access.

9 scopes. Give each integration exactly the permissions it needs.

passports:read

List and retrieve passports

passports:write

Create new passports

verify

Verify intent envelopes

revoke

Revoke or suspend agents

analytics:read

Read verification analytics

keys:manage

Create & rotate API keys

billing:read

View plan & usage

webhooks:manage

Configure webhook endpoints

admin

Full account access

Start verifying agents today.

Free tier includes 100 verifications/day, 5 agent passports, and full CLI access. No credit card required.

or just pip install aip-protocol — the SDK works fully offline too