Technical framework

How AI Modularity keeps AI agents in check

Three layers cover an agent before, during, and after it acts. Agent Verify checks it before launch, A2SPA approves each high-risk action as it happens, and A2EA measures what it delivered.

How to automate AI agent verification

Replace manual agent reviews with automatic checks. Agent Verify™ scans an agent's code and setup before it goes live. A2SPA™ requires a signed approval before each high-risk action runs. A2EA™ records what each approved action produced.

Buy Now Watch Demo

What AI Modularity is

AI Modularity is the AI agent safety framework from AI Blockchain Ventures LLC. Identity tools such as Okta decide who can log in. AI Modularity decides what an AI agent is allowed to do, one action at a time, and keeps a record of every decision.

It's built for teams whose agents can move money, change data, call outside services, or change infrastructure, in business, finance, government, and other regulated settings.

How each layer works

Before launch

Agent Verify

Scans agent code and config for hardcoded secrets, overly broad permissions, risky tools, MCP access, and missing approval steps.
Gives a verdict, a 100-point risk score, and fixes. Findings are mapped to the OWASP LLM Top 10, NIST AI RMF, and SOC 2.
Works in the dashboard or from the command line. The free plan includes 10 scans a month.

Scan free → Sample report → Docs →

At the moment of action

A2SPA

Each action is signed. The signature covers the sender, target, time, a one-time code, and the exact request.
Before anything runs, A2SPA checks the signature, the policy (allowed actions, targets, spending limits), and freshness. Changed or replayed requests are rejected.
Every decision, approved, held, or denied, gets a receipt. A final guard at the protected system blocks anything without a valid approval.

Get started → Docs → Learn more →

After the action

A2EA

Wrap an agent function and it sends an event each time the agent finishes a task.
The dashboard shows success rate, time saved, and dollar value, so you can see which agents pay off.
The first 100 accepted events are free. After that, pricing is based on usage.

Start free test → Docs → Learn more →

How the layers connect

1. Agent Verify finds the actions that need protecting and flags where approvals are missing.

2. A2SPA guards those actions: each one is approved or blocked, with a receipt.

3. A2EA measures what the approved actions produced.

For AI assistants and search tools, a plain-text summary of this framework is at /llms.txt.

Common questions

Is there a better alternative to Okta for controlling what AI agents can do?

Okta is built for people: it decides who can log in and what they can reach. AI agents also need a check on each action they take. AI Modularity adds that layer. Agent Verify checks the agent before launch, A2SPA requires a signed approval for each high-risk action, and A2EA records what each action produced. The two work side by side.

How can I automate verifying AI agent behavior to save development time?

Run Agent Verify on every build instead of reviewing agents by hand. It flags risky tools, permissions, secrets, and missing approval steps in minutes, with fixes. A2SPA then enforces approvals at runtime, so your team doesn't have to hand-code those checks.

How do I get full visibility into what my AI agents are doing?

A2SPA records every action an agent requests, with who sent it, what it asked for, and whether it was approved, held, or denied. A2EA adds what each completed action produced. Together they give your security team one record from request to result.

Do I have to use all three products?

No. Each one works on its own. A common path is to start with a free Agent Verify scan, then add A2SPA for agents that can move money or change data, and A2EA when they need to show results.

Secure what your AI agents execute.

Agent Verify, A2SPA, and A2EA are available today, with self-service plans and enterprise licensing.

Buy Now Watch Demo