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

> ## Agent Instructions
> To set up Hue in an application, fetch https://docs.hue.run/guides/agent-setup.md and follow it.
> Hue's project-data MCP server is https://mcp.hue.run/mcp. https://docs.hue.run/mcp only searches these docs and cannot read a Hue project.
> Never ask the user to paste a Hue API key into chat; they store it as HUE_API_KEY themselves.
> Install the exact SDK versions pinned on the page you're following, or on https://docs.hue.run/quickstart.md. Run the CLI as `npx @hue-run/sdk` or `bunx @hue-run/sdk`, never `npx hue`.
> Use Hue product terms: eval set, case, evaluator, run and scoring run. Older APIs may say dataset, scorer or experiment; prefer the current SDK methods. See https://docs.hue.run/concepts.md.

# Concepts

> The objects Hue records and evaluates, where to find them in the app, and their names in the SDKs and MCP.

Hue observes your application through OpenTelemetry traces, then turns traces into cases you evaluate repeatedly.

## Observe

| Term | What it is | In Hue |
| - | - | - |
| Trace | One recorded execution of your application: a tree of OpenTelemetry spans with one root. | **Traces** |
| Span | One timed step in a trace, such as a model call, tool call or agent step. | Trace detail |
| Session | Traces grouped by the session id your application records, such as turns in one conversation. | **Traces** filter |
| Finding | A problem Hue detects in an ended trace, such as an unrecovered tool error or a reply that claims an action no tool performed. A trace with a finding needs attention. | [**Traces**](/agents/mcp-tools#attention-states-and-findings) |
| Trace check | A project-defined, versioned check of production behavior, such as whether a handoff expanded the request. Advisory, not a score. | [**Traces**](/agents/mcp-tools#trace-checks) |
| Intent | The task type a trace belongs to, classified against your project's published taxonomy. | [**User Intent**](/agents/mcp-tools#intents) |

## Evaluate

| Term | What it is | In Hue |
| - | - | - |
| Eval set | A versioned collection of cases. A run pins one eval set version. | **Evals** |
| Case | One example: the task, its inputs, optional expected values and reviewed success criteria. You can create one from a trace. | [**Evals**](/evaluations/case-from-trace) |
| Evaluator | How quality is measured: a deterministic **verifier** or a hosted LLM **judge** with a pinned model and rubric. | [**Evaluators**](/evaluations/first-evaluation) |
| Run | Your agent executed over an eval set version, with each case's output, trace and scores. Compare runs to catch regressions. | [**Runs**](/evaluations/first-evaluation) |
| Scoring run | Evaluators applied to outputs a run already stored, without running the agent again. | [**Runs**](/evaluations/first-evaluation) |
| Environment | A simulated app (for example Linear or Gmail): its starting data and the actions an agent can take. | [**Environments**](/evaluations/simulations) |
| World | A fresh instance of an environment version for one execution. Its journal of actions is scoring evidence. | [**Environments**](/evaluations/simulations) |

## Access

| Term | What it is |
| - | - |
| Project | The workspace that holds traces, eval sets, runs and keys. |
| Project key | A server-side key with one preset: **Read**, **Read and write** or **Tracing only**. Created in **Settings → Integrations & API keys**. See [Project keys](/quickstart#project-keys). |
| Hue MCP | The project-data MCP server at `https://mcp.hue.run/mcp`. Agents connect by signing in with Hue or with a key. See [Hue MCP](/agents/mcp-server). |

## Names in code

The SDK evaluation clients and MCP tools use these product names: `createEvalSet`, `createEvaluator` and `createRun` in TypeScript, `create_eval_set`, `create_evaluator` and `create_run` in Python, and fields such as `eval_set_id`, `evaluator_version_id`, `run_id` and `scoring_run_id`. Older methods, REST paths and response fields still say dataset, scorer and experiment; the [TypeScript](/reference/typescript) and [Python](/reference/python) references mark which ones are deprecated.


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