> For the complete documentation index, see [llms.txt](https://stair-ai.gitbook.io/stair-ai-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://stair-ai.gitbook.io/stair-ai-docs/readme/introduction.md).

# What is Stair AI

This page expands the thesis the [landing page](/stair-ai-docs/readme.md) sketches: why the trust gap exists, what fills it, and where each piece sits in the architecture.

## Before Stair AI

An AI agent's trustworthiness is whatever it claims about itself. Predictions arrive without provenance. Reasoning is locked inside the agent's memory, instrumented as fragments of mutable text, never persisted in a form anyone else can audit. Track records can be edited after the fact: keep the wins, drop the losses. Identity is a wallet, and wallets are disposable.

The trust gap is industry-agnostic. Concrete scenarios across Web3 and traditional domains live on [The problem](/stair-ai-docs/readme/problem.md).

## After Stair AI

Glass Box Protocol turns AI reasoning into an append-only, replayable record. Every agent action is preceded by a structured representation of why: the trigger, the data consumed, the inferences drawn, the alternatives considered, the confidence claimed. Each record is captured under a typed schema and stored in custody held independently of the agent's developer, with the cryptographic hash of its predecessor included so tampering is detectable end-to-end without any external anchor.

A scoring layer runs on top and evaluates the reasoning, not the outcome: process quality independent of whether the agent's prediction was correct. The result is a credit-score equivalent for autonomous agents, anchored on how the agent thinks rather than on who controls it — auditable from any DeFi protocol, DAO, prediction market, insurer, or regulator, without trusting Stair AI as an intermediary.

```mermaid
%%{init: {"layout": "elk"}}%%
graph TD
  %% --- Skill Layer ---
  subgraph Skill["Skill layer"]
    direction TB
    Data[<strong>Data Sources</strong><br/>e.g. volatility index, BTC whale activities stream]
    Tools[<strong>Tools</strong><br/>e.g. API calls]
    Logic[<strong>Logic</strong><br/>e.g. trading strategy logic, risk framework, technical analysis model]
    Module[<strong>Skill Module</strong><br/>defines input, reasoning logic, constraints, and output format]
    Data --> Module
    Tools --> Module
    Logic --> Module
  end

  %% --- Intelligence Layer ---
  subgraph Intel["Intelligence layer"]
    direction LR
    Eval[<strong>Eval Engine</strong><br/>Backtesting and simulated environments for scoring]
    Ledger[<strong>Reasoning Ledger</strong><br/>Skill identity, reasoning trace, confidence scores, tools/data sources]
    Scoring[<strong>Scoring Engine</strong><br/>Accuracy/Quality scoring, skill reliability + robustness]
  end

  %% --- Application Layer ---
  subgraph App["Application layer"]
    direction LR
    Ranking[<strong>Skill Ranking</strong><br/>Performance metrics for multiple use case categories]
    API[<strong>Intelligence API</strong><br/>Dev tool for querying skill performance and subscribing to skills]
    UseCases[<strong>Agent Finance Ecosystem Use Cases</strong><br/>E.g. AI bank, smart vaults, agent lending, trading agents,<br/>yield farming agents, DAO treasuries]
    Ranking --> UseCases
    API --> UseCases
  end

  %% --- Cross-layer connections ---
  Module --> Eval
  Module --> Ledger
  Module --> Scoring

  Eval --> Ranking
  Ledger --> API
  Scoring --> Ranking

  %% --- Styling ---
  classDef skill fill:#fff8c4,stroke:#caa500,color:#000
  classDef intel fill:#d4e3f5,stroke:#5a7ca8,color:#000
  classDef app fill:#ffd9a8,stroke:#cc7733,color:#000

  class Data,Tools,Logic,Module skill
  class Eval,Ledger,Scoring intel
  class Ranking,API,UseCases app
```

The protocol decomposes into three layers. The Skill layer is agent-side: data sources, tools, and strategy logic compose into a Skill Module that defines the agent's reasoning contract. The Intelligence layer is what Stair AI runs: an Eval Engine that backtests skills in simulation, a Reasoning Ledger that captures live traces under independent custody, and a Scoring Engine that grades reasoning quality independent of outcome. The Application layer is what consumers integrate: a Skill Ranking surfacing performance across use-case categories, an Intelligence API for querying ledger contents, and the downstream ecosystem of AI banks, smart vaults, lending markets, trading and yield farming agents, and DAO treasuries. Yellow marks components contributed by skill developers; blue, Stair protocol infrastructure; orange, consumer-facing applications.

Cryptographic anchoring on a public chain is the upgrade path for agent-to-agent verification, where two autonomous agents need to verify each other's records without a human intermediary. Until then, immutability and replayability come from the append-only schema and independent custody.

## Read next

* [The problem](/stair-ai-docs/readme/problem.md). The trust gap in concrete terms, in Web3 and outside it.
* [The five-layer problem stack](/stair-ai-docs/readme/problem-stack.md). Why this is hard, in order.
* [Where Stair AI sits](/stair-ai-docs/readme/positioning.md). The protocol's place in the agent infrastructure stack.
* [Where this applies](/stair-ai-docs/readme/use-cases.md). Six concrete use cases across industries.
* [Design principles](/stair-ai-docs/readme/design-principles.md). The six beliefs that frame every decision.
* [Project status](/stair-ai-docs/readme/status.md). What is shipped today and what is next.