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

# Core concepts

> The small set of concepts used throughout Failproof AI.

| Concept | What it means | What you can achieve |
| - | - | - |
| Session | One agent task or run | Reconstruct an individual outcome |
| Event | One recorded action in a session | Inspect a model call, tool use, error, human action, or policy decision |
| Trace | The ordered and nested session view | Understand causality instead of reading disconnected logs |
| Evaluation | A score or judgment on a session | Track quality, compliance, cost, or latency continuously |
| Audit | A review of a selected session population | Search for failure patterns with a stated goal and cadence |
| Finding | An evidence-backed failure discovered by an audit | See the failure mode, severity, and affected sessions |
| Issue | A durable response record | Assign, discuss, and resolve a finding or alert incident |
| Alert | A rule that detects recurrence | Notify responders when a known condition returns |
| Policy | A rule evaluated during agent activity | Observe, block, or redirect risky behavior |
| Deployment | A versioned policy rollout to machines | Control where a policy runs and roll it back safely |

## The reliability loop

Start from evidence. A session shows what happened. An audit determines whether it is an isolated mistake or a pattern. A finding identifies the failure mode; an issue owns the response. A policy prevents the same behavior, while alerts tell you if the condition returns.

<Warning>
  A local `failproofai audit` scans local agent history. A recurring Cloud audit reviews sessions stored in Failproof AI Cloud. They are separate workflows with different scope and scheduling.
</Warning>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.