The trust layer for AI agents

Give your agent limits. Prove every action it takes.

AION gives AI agents what banks gave humans: a signed mandate (budget + payee allowlist), one-time payment authorizations bound to an exact amount, hash-chained receipts with externally anchorable roots, and court-ready dispute bundles — evidence anyone can verify with math. Scan code, guard dangerous actions, and enforce human intent on real x402 payments.

Install from PyPI See Demo GitHub

$ pip install aion-core
$ aion scan .
HIGH shell command execution found
HIGH file deletion capability found
MEDIUM MCP config reference found

$ aion guard-demo
ALLOW file.read in 0.1ms
BLOCK shell.run rm -rf
APPROVAL shell.run echo hello

$ aion receipts 3
proof created for every action

The first AION workflow

AION starts with a simple loop: scan the project, guard risky actions, and keep receipts.

AION Scan

Find risky patterns like shell commands, file deletion, secret access, database writes, network calls, and MCP configs.

AION Guard

Allow safe actions instantly, require approval for high-risk actions, and block forbidden actions.

AION Receipts

Create proof for important actions: agent, scope, decision, risk, status, timestamp, and receipt hash.

Install and test in five minutes

Install, run `aion doctor` to verify, `aion demo` to see the full trust layer, then inspect receipts.

Step 1

Install package

pip install aion-core

Step 2

Scan a project

aion scan .

Step 3

Run the guard demo

aion guard-demo

Step 4

View receipts

aion receipts 10

Proof from the current build

AION Guard allowed low-risk file.read in about 0.1ms. Focused Guard + Scan tests pass locally.

Action type	AION behavior
Low-risk	Allow instantly
Medium-risk	Allow and create receipt
High-risk	Require approval
Forbidden	Block instantly