> ## Documentation Index
> Fetch the complete documentation index at: https://docs.manthan.systems/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> What Parmana is, the problem it solves, and how it fits into an AI system

## What Parmana is

Parmana is **Execution Trust Infrastructure** for AI systems. It sits between an AI system's decisioning and the execution of that decision, and answers one question for every action an AI system wants to take:

> Is this action authorized, and can that authorization be independently proven?

Parmana is not an AI orchestration framework, and it is not itself AI-powered. It is a deterministic rule-enforcement layer: given the same policy and the same inputs, it always produces the same authorization decision. Parmana **enforces** authority that humans have already defined — it does not decide on their behalf, and it has no human-in-the-loop escalation step at runtime. The model is strictly allow or block.

## The problem

AI systems can generate decisions and execute actions, but most organizations cannot answer basic questions about those actions after the fact:

* Was this action authorized?
* Which policy approved it?
* Which business intent was evaluated?
* What evidence supports the execution?
* Can the execution be independently verified, without trusting the system that produced it?

Without answers to these questions, AI cannot safely operate in regulated or mission-critical workflows.

## The solution

Parmana introduces an execution authorization layer between AI systems and the systems that actually carry out actions.

1. An AI system produces an **Intent** — a requested business action.
2. Parmana evaluates that Intent against an immutable, versioned **Policy**.
3. Parmana produces a deterministic **Decision**: approved or rejected.
4. Only approved actions are permitted to **Execute**.
5. Execution produces immutable **Evidence**, which is aggregated into an **Execution Trust Record**.
6. Any party can independently **Verify** that record later, without needing access to the original runtime.

```
AI System
    |
    v
AI Intent
    |
    v
Parmana Authorization Runtime
    |
    +-- Load Policy
    +-- Evaluate Policy
    +-- Create Decision
    |
    v
Execution Authorization (Approved / Rejected)
    |
    v
Execution System
    |
    v
Execution Evidence
    |
    v
Execution Trust Record
    |
    v
Independent Verification
```

Only Parmana-approved actions are executed.

## Key guarantees

* AI executes only policy-compliant actions.
* Every execution is explicitly authorized before it runs.
* Policy evaluation is deterministic — same policy, same inputs, same decision.
* Every execution produces immutable evidence.
* Every Execution Trust Record is cryptographically verifiable.
* Every execution can be independently audited without trusting the original runtime.
* Policy evaluation and execution are cleanly separated responsibilities.

## Where to go next

<CardGroup cols={2}>
  <Card title="Installation" href="/docs/getting-started/installation" icon="download">
    Set up the monorepo, TypeScript SDK, or Python SDK locally.
  </Card>

  <Card title="Quickstart" href="/docs/getting-started/quickstart" icon="bolt">
    Run your first Business Transaction end to end.
  </Card>

  <Card title="Concepts" href="/docs/concepts/overview" icon="brain">
    Learn the domain model: Authority, Intent, Policy, Decision, Execution, Evidence.
  </Card>

  <Card title="Architecture" href="/docs/architecture/overview" icon="sitemap">
    See how the Runtime, packages, and trust chain fit together.
  </Card>
</CardGroup>