See what your AI agent really does.
Aptha Assure (Agent Evaluator)
Aptha Assure evaluates supply-chain AI agents, to show whether they are safe and correct enough for production, with evidence you can inspect.
Aptha Analyzer
The Analyzer reads your agent's code against its Agent Functional Spec and reports what the agent decides, which limits it really enforces and which systems it touches. Where code and spec disagree, it points at the line.
What you give it
An Agent Functional Spec (JSON)
What the agent is for, the decisions it makes, its limits and the systems it uses.
No spec yet? The Spec toolkit has a prompt that writes it from your source, and a checker.
The agent's source code
A public
https:// Git URL, or a ZIP (use a ZIP for private code).Python is best supported; JavaScript, TypeScript, Java and Go get lighter checks.
What you get back
Workflow graph
What the spec declares and what the code does, side by side. Each box is coloured by whether the code backs it up.
Verification checks
Every guardrail, formula and authority level you declared: verified, mismatch, not found or unverifiable, with file and line.
Quick guide
- 1Create an account
Your analyses stay private to you.
- 2Prepare the spec
Use the Spec toolkit: a prompt for your coding agent writes it from your source.
- 3Analyze
Give the spec and the source (public Git URL or ZIP). It takes a minute or two.
- 4Read the results
Overview, Workflow graph (with the JSON download), Verification & evidence.
Your spec and results are visible only to your account. Source code is deleted as soon as the analysis finishes.
The Analyzer reads code; it does not run your agent, and it is not a certification. “Unverifiable” means the code alone cannot settle a claim, not that it is wrong.
Want to see a finished analysis first?
Open a public example of an open-source agent. No account needed.