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

> ## Agent Instructions
> These docs moved from docs.flintai.dev to docs.switchagents.ai. Use docs.switchagents.ai for every link and request.
> To search these docs from an AI tool, connect the MCP server at https://docs.switchagents.ai/mcp. The page index is at https://docs.switchagents.ai/llms.txt.
> The source code for these products is public on GitHub. Switch Rooms: https://github.com/sandbox-quantum/switch. Switch Trust CLI: https://github.com/sandbox-quantum/switch-trust-cli. Switch Trust CodeScan action: https://github.com/sandbox-quantum/switch-trust-codescan-action. Switch Trust MCP server: https://github.com/sandbox-quantum/switch-trust-mcp. Switch Trust Python SDK: https://github.com/sandbox-quantum/switch-trust-sdk-py. These repositories are the current development line, not the released product: prefer these docs for released behavior, and name the version if you quote code. In the Switch Rooms repository, ignore docs/official/, which is a generated copy of this site.

# Glossary

> Switch Trust and Switch ROI terms and what they mean

Terms used across the Switch Trust and Switch ROI docs, in alphabetical order.

## Agent

An AI agent is software that uses AI to carry out tasks on a user's behalf, often using tools and making decisions about how to complete the work.

In Switch Trust, an agent is a tracked record discovered by a scan or reported by a connected [data source](#data-source). Each record may represent an individual agent or a group, depending on the source. For example, the [Claude Code plugin](#claude-code-plugin) creates a separate agent for each person who installs it, while Switch Rooms groups all rooms on a Switch server into a single agent named **Switch Rooms Agents**.

## Agent evaluation

An agent evaluation tests your agent by sending prompts to its endpoint and scoring the responses. The results combine into the agent's [overall evaluation health](#overall-evaluation-health).

Evaluations can test the agent with adversarial prompts, check the quality of its answers, or score it with a judge prompt you write. Each result is either a **Pass** or **Fail**, or a score from 0 to 100. An agent evaluation is different from a [model evaluation](#model-evaluation), which tests the model rather than your agent. See [Evaluate an agent](https://docs.switchagents.ai/switch-trust/getting-started/evaluate.md).

## API key

An API key lets a tool or script connect to your Switch Trust workspace. Each key has a [role](#role) that limits what it can do.

An **Ingest only** key can send data and read guardrail policies, but can't read your data. That's enough for the SDK, a data source, or the Claude Code plugin. Some setup steps call an API key a token. See [API keys](https://docs.switchagents.ai/switch-trust/admin/api-keys.md).

## Asset

An asset is something an agent uses: a model, an MCP server, or a tool. Agents aren't assets.

Some labels in the product, such as **Assets analyzed**, count agents as well. See [Assets](https://docs.switchagents.ai/switch-trust/inventory/assets.md).

## Burn rate

Burn rate is an agent's average daily spend this month: its spend so far divided by the day of the month, in UTC.

For an agent that started reporting mid-month, burn rate reads low for the rest of that first month. See [Spending and caps](https://docs.switchagents.ai/switch-trust/cost/spending-and-caps.md).

## Cap

A cap is a monthly spending limit you set on an agent. Switch Trust forecasts when the agent will reach it, and marks the agent **Over cap** once spending passes it.

A cap doesn't stop the agent from spending, and it doesn't send an alert. See [Spending and caps](https://docs.switchagents.ai/switch-trust/cost/spending-and-caps.md).

## Claude Code plugin

The Claude Code plugin reports each person's Claude Code sessions and cost to Switch Trust, with no code changes.

The plugin only reports activity. It doesn't check or block anything. See [Claude Code plugin](https://docs.switchagents.ai/switch-trust/admin/data-sources/claude-code.md).

## Data source

A data source is a way to bring agents into Switch Trust.

Data sources include code repositories and cloud accounts to scan, OpenTelemetry, the Claude Code plugin, and Switch servers. See [Data sources](https://docs.switchagents.ai/switch-trust/admin/data-sources/index.md).

## Detector

A detector is a single check in a [guardrail policy](#guardrail-policy), such as a check for jailbreaks, toxicity, or credit card numbers.

Each detector checks the side of the conversation its [detector role](#detector-role) sets, and takes its [guardrail action](#guardrail-action) when it finds a match. See [Guardrail policies](https://docs.switchagents.ai/switch-trust/guardrails/policies.md).

## Detector role

A detector role sets which side of a conversation a detector checks:

* **User:** Checks what goes into the agent, such as prompts.
* **Assistant:** Checks what comes out, such as the model's replies.

In Switch Rooms, **User** checks messages from people and **Assistant** checks messages from agents.

A detector checks only the roles it's set on. Some detectors work on one side only: jailbreak and topic control check only **User**, and tool permissions check only **Assistant**. So in Switch Rooms, a jailbreak detector never checks messages from agents. A detector role is different from a [role](#role), which sets what a person can do.

## Discovery

Discovery scans your code repositories and cloud accounts for AI agents and the [assets](#asset) they use, and adds what it finds to your [inventory](#inventory).

A repository scan finds agents in Python code. In JavaScript and TypeScript, it recognizes only the Anthropic, OpenAI, Claude Agent, and MCP SDKs, and finds agents only from the Claude Agent SDK. Each repository needs the scan workflow added, and an API key from your own LLM provider. Scans don't count toward [managed agents](#managed-agent). See [Find your agents](https://docs.switchagents.ai/switch-trust/getting-started/discover.md).

## Guardrail

A guardrail checks an agent's messages or requests against a [guardrail policy](#guardrail-policy) and takes action when it detects a violation.

Guardrails apply to traffic routed through the Switch Trust [guardrails gateway](#guardrails-gateway), including traffic from agents using the [SDK](#sdk), and to Switch Rooms messages. OpenTelemetry and the Claude Code plugin only report activity to Switch Trust; guardrails do not apply to traffic reported through these sources.

## Guardrail action

A guardrail action determines what happens when a [detector](#detector) finds content that matches its criteria:

* **Alert:** Allows the content through and records the match.
* **Redact:** Replaces the matched text before allowing the content through.
* **Block:** Prevents the content from passing through.

Only **Redact** and **Block** change what gets through.

## Guardrail policy

A guardrail policy defines what guardrails check and how they respond. It contains a set of [detectors](#detector), each configured with a [detector role](#detector-role) and a [guardrail action](#guardrail-action).

An agent or Switch server selects a policy by sending its **Policy ID**. There is no default policy, so no guardrail checks run unless a Policy ID is provided. See [Guardrail policies](https://docs.switchagents.ai/switch-trust/guardrails/policies.md).

## Guardrails gateway

The guardrails gateway is the Switch Trust service that checks an SDK agent's model traffic. It checks each request, and the response that comes back, against the guardrail policy the request names, then passes them on.

The SDK reaches the gateway at the **gateway URL**. The gateway doesn't check traffic from an agent that's paused, past the limit on your plan, or sending no agent name. See [managed agent](#managed-agent) and [Configuration](https://docs.switchagents.ai/switch-trust/sdk/python/configuration.md#gateway-url).

## Improvements

Improvements are Switch ROI recommendations for lowering an agent's cost, such as moving work to a smaller model.

Each improvement has a confidence level, and most have an estimated savings. Improvements are based on the sessions the agent has recorded, so a new agent's improvements start with low confidence. See [Improvements](https://docs.switchagents.ai/switch-trust/inventory/agents.md#improvements).

## Inventory

Your inventory is every agent Switch Trust knows about, with the [assets](#asset) each one uses.

[Discovery](#discovery) adds agents from scans, and data sources add agents as they report. See [Agents](https://docs.switchagents.ai/switch-trust/inventory/agents.md).

## Issue

An issue is a problem Switch Trust found in your code, your cloud, or your agents' live traffic.

Switch Trust groups problems into issues by [rule](https://docs.switchagents.ai/switch-trust/rules/index.md) and [severity](#severity), so one issue can have many [occurrences](#occurrence). See [Discovery results](https://docs.switchagents.ai/switch-trust/discovery/results.md).

## Managed agent

A managed agent is one Switch Trust guards or evaluates. An agent becomes managed the first time a guardrail checks its traffic or an evaluation run finishes for it. Sessions and cost from OpenTelemetry or the Claude Code plugin don't make it managed, and neither does a scan.

Your plan sets how many managed agents you can have. At the limit, a new agent's guardrails aren't applied, so its traffic reaches the model unchecked, and its evaluations don't run. An Org Admin can pause an agent's management to free its place. See [Plans and billing](https://docs.switchagents.ai/switch-trust/admin/plans-and-billing.md#what-counts-as-a-managed-agent).

## MCP server

An MCP server gives an agent tools and data through the Model Context Protocol (MCP), an open standard for connecting agents to the services they use.

In Switch Trust, MCP servers are [assets](#asset) in your inventory. Switch Trust also has its own MCP server, which brings an agent's cost, its improvements, and the issues in your code into your coding assistant. See [Optimize with the MCP server](https://docs.switchagents.ai/switch-trust/mcp/index.md).

## Model evaluation

A model evaluation tests a model your agents use, such as for jailbreaks, misuse, and toxicity. SandboxAQ runs it, and the result is that model's [model health score](#model-health-score).

It tests the model on its own, without the instructions or guardrails you give your agent, so the result is the same for every customer whose agents use that model. You don't run it. To test how your own agent behaves, see [agent evaluation](#agent-evaluation).

## Model health score

A model health score rates how a model behaves under testing, based on its [model evaluation](#model-evaluation). It comes with sub-scores for jailbreaks, misuse, toxicity, security, and robustness.

It describes the model, not your agent. The score for your agent is its [overall evaluation health](#overall-evaluation-health). See [Evaluation results](https://docs.switchagents.ai/switch-trust/evaluation/results.md).

## Model pricing

Model pricing is the per-token rate Switch Trust uses to calculate cost: the list price, or a negotiated rate you set. Cost figures are estimates at those rates, not your bill.

A model with no price adds nothing to totals, and a session that uses it shows no cost, until you set a rate. See [Set model pricing](https://docs.switchagents.ai/switch-trust/cost/model-pricing.md).

## Occurrence

An occurrence is one agent or asset where a [rule](https://docs.switchagents.ai/switch-trust/rules/index.md) matched.

Occurrences with the same rule and [severity](#severity) group into one [issue](#issue), and the **Issues** table counts them under **Occurrences**. Some pages call an occurrence a finding.

## OpenTelemetry

OpenTelemetry is an open standard for collecting data about what software is doing, such as a trace of each request and measurements like timing, and sending it to a tool that stores and analyzes it.

In Switch Trust, OpenTelemetry is a [data source](#data-source) that sends an agent's sessions and cost, set up with environment variables. It doesn't apply guardrails, and it doesn't make an agent managed. See [OpenTelemetry](https://docs.switchagents.ai/switch-trust/admin/data-sources/opentelemetry.md).

## Organization

Your organization is the space for your company in Switch Trust. It holds the members and their roles, the plan, and the count of [managed agents](#managed-agent).

A new person joins by invitation from someone already in the organization. Signing up without an invitation creates a new organization. You can belong to more than one organization and switch between them. See [Organization](https://docs.switchagents.ai/switch-trust/admin/organization.md).

## Overall evaluation health

Overall evaluation health is a single score for your agent: the weighted average of the [agent evaluations](#agent-evaluation) assigned to it.

It's different from the [model health score](#model-health-score), which rates the model your agent uses. See [Agent results](https://docs.switchagents.ai/switch-trust/evaluation/agent-results.md).

## PII (personally identifiable information) and PHI (protected health information)

Personally identifiable information (PII) is data that can be traced to a specific person, such as a name, email address, or phone number. Protected health information (PHI) is health data about a person, along with details that identify them, such as names and addresses.

In Switch Trust, the PII detectors find credit card numbers, email addresses, phone numbers, Social Security numbers, and IP addresses, and can redact or block them. The PHI detector can only alert. See [Guardrail policies](https://docs.switchagents.ai/switch-trust/guardrails/policies.md).

## Role

A role sets what a person or an [API key](#api-key) can do in Switch Trust, from **Viewer** to **Org Admin**.

A role is different from a [detector role](#detector-role), which sets which side of a conversation a detector checks. See [Users](https://docs.switchagents.ai/switch-trust/admin/users.md#roles-and-permissions).

## SDK

An SDK, or software development kit, is a library you add to your own code.

The Switch Trust SDK wraps the LLM client in your agent so its traffic goes through Switch Trust. It's how you apply guardrails to an agent in your own code, and it also reports sessions and cost. It's available for Python and TypeScript. See [Monitor and protect with the SDK](https://docs.switchagents.ai/switch-trust/getting-started/runtime.md).

## Session

A session is a group of [turns](#turn) from one conversation or task.

In Switch Rooms, each room is one session.

## Severity

Severity is how serious an [issue](#issue) is: Critical, High, Medium, Low, or Informational.

The rule that finds an issue sets its severity. See [How discovery works](https://docs.switchagents.ai/switch-trust/discovery/how-discovery-works.md).

## Switch ROI

Switch ROI is the part of Switch Trust that shows what each agent costs and how to lower it. See [Switch ROI](https://docs.switchagents.ai/switch-trust/cost/index.md).

## Turn

A turn is one prompt and the response that comes back.

In Switch Rooms, each checked message is one turn, with no response. Some pages and API fields call a turn an interaction.

## Workspace

A workspace holds your agents, sessions, cost, guardrail policies, and evaluations.

Each organization has one workspace, and every API key belongs to that workspace. See [Organization](https://docs.switchagents.ai/switch-trust/admin/organization.md).
