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

# BotGauge documentation

> Find out how your AI agent fails before your users do. Red-team it, trace it, and manage your team with BotGauge. Start with a 5 to 15 minute quickstart.

BotGauge finds out how your AI agent fails before your users do. It attacks the agent the way a real attacker would, records the full execution path behind every result, and lets you replay a finding to confirm a fix.

[Get started free →](https://cloud.botgauge.com)
[Book a demo](https://calendly.com/botgauge/30min)

## Choose your quickstart

> **[Red-team an agent](https://www.botgauge.com/docs/red-teaming/quickstart.md)**
>
> **Best for:** teams that want to know how an agent breaks under attack.
>
> Connect an endpoint, approve a plan, and read the results. About 15 minutes, with a sample agent included.

> **[Trace an agent](https://www.botgauge.com/docs/tracing/quickstart.md)**
>
> **Best for:** teams that want to see every prompt, tool call, and span behind a result.
>
> Use the Python SDK or any OpenTelemetry exporter. About 5 minutes.

You can start with either one. Red teaming needs only your agent's endpoint, and tracing needs the SDK or an exporter in your app.

## How the pieces fit together

A red-team run tells you what broke. Tracing shows how the agent got there. Once you've fixed the agent, replaying the attack confirms the fix.

```mermaid theme={null}
flowchart LR
  A[Red team your agent] --> B[Findings with evidence]
  B --> C[Trace the execution path]
  C --> D[Fix the agent]
  D --> E[Verify fix: replay the attack]
  E --> A
```

## Find your way

| If you are | Start here |
| - | - |
| A developer shipping an agent | [Tracing quickstart](https://www.botgauge.com/docs/tracing/quickstart.md), then your [framework](https://www.botgauge.com/docs/integrations/overview.md) |
| On a security or AI safety team | [Red teaming overview](https://www.botgauge.com/docs/red-teaming/overview.md), then [Read your results](https://www.botgauge.com/docs/red-teaming/results.md) |
| Working on compliance or risk | [Compliance view](https://www.botgauge.com/docs/red-teaming/compliance.md) |
| Setting up your team | [Setup](https://www.botgauge.com/docs/get-started/setup.md), then [Platform concepts](https://www.botgauge.com/docs/platform/concepts.md) |
| New to the terms | [Core concepts](https://www.botgauge.com/docs/get-started/core-concepts.md) |

## Explore the docs

> **[Get started](https://www.botgauge.com/docs/get-started/core-concepts.md)**
>
> Core concepts and account setup.

> **[Red teaming](https://www.botgauge.com/docs/red-teaming/overview.md)**
>
> Connect, recon, plan, run, and read results.

> **[Tracing](https://www.botgauge.com/docs/tracing/overview.md)**
>
> Sessions, turns, agents, and spans over OpenTelemetry.

> **[Integrations](https://www.botgauge.com/docs/integrations/overview.md)**
>
> Eleven frameworks, plus any OTLP exporter.

> **[Platform](https://www.botgauge.com/docs/platform/concepts.md)**
>
> Organizations, workspaces, projects, roles, keys, and secrets.

> **[Reference](https://www.botgauge.com/docs/reference/cli.md)**
>
> CLI commands and environment variables.

## Common questions

> **Do I need to change my agent's code?**
>
> For red teaming, no. BotGauge needs only your agent's endpoint, and a small optional wrapper unlocks deeper tests. See [black box and grey box](https://www.botgauge.com/docs/red-teaming/black-box-and-grey-box.md). For tracing, the [zero-code launcher](https://www.botgauge.com/docs/tracing/zero-code.md) traces a Python app without edits.

> **Is it safe to run against my agent?**
>
> BotGauge never attacks until you approve a specific plan, it acts only within the scope you approved, and it refuses real actions in production. See [Approve and run](https://www.botgauge.com/docs/red-teaming/approve-and-run.md).

> **Which frameworks are supported?**
>
> LangChain, LangGraph, OpenAI SDK, Anthropic SDK, LlamaIndex, CrewAI, OpenAI Agents, Google ADK, AutoGen AgentChat, PydanticAI, and MCP. Any other OpenTelemetry source can send traces through [OTLP](https://www.botgauge.com/docs/integrations/otlp.md).

> **Does a clean report mean my agent is secure?**
>
> No. It means the tested rules held under the attacks tried. Only a result of Held is a pass, and a rule that couldn't be tested is a gap, not safety. See [Read your results](https://www.botgauge.com/docs/red-teaming/results.md).

> **Where do my keys and secrets live?**
>
> Tracing uses [project API keys](https://www.botgauge.com/docs/platform/api-keys.md). Red teaming reads credentials your agent needs from [project secrets](https://www.botgauge.com/docs/platform/secrets.md), and BotGauge never shows the values back.

> **[See BotGauge in action](https://calendly.com/botgauge/30min)**
>
> Get a 30-minute walkthrough of red teaming and tracing for your agent.

> [!NOTE]
>
> The tracing SDK is an early release (0.x). Public API shapes may still change before 1.0.


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