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

# Agent View

**Agent View** brings together everything about a single agent: its configuration, the context and tools it runs on, its sessions, traces and logs, and how its outputs are reviewed and scored.

Use Agent View to:

* Check the agent's configuration and what changed between revisions
* See which context items and tools the agent uses
* Review its traces and sessions
* Test it before and after a change

### Overview

The **Overview** tab shows how the agent is doing over the selected date range, with each metric compared against the previous period. Sessions, traces, observations, and scores mean the same as on Home, scoped to this agent. Two metrics are specific to Agent View:

* **Avg. traces per session**: how many requests a conversation takes, on average. A rise can mean customers need more back-and-forth to get what they came for.
* **Avg. iterations per trace**: how many reasoning steps the agent takes per request. A rise can mean policies or routines are making each decision harder.

<div data-with-frame="true"><figure><img src="https://708770081-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1ICwJbq7EJdn5kBgXnQu%2Fuploads%2FAx1hYvAkrGxHab55lH2P%2Fpo-matching-agent.gif?alt=media&amp;token=137393fe-fcc1-41c3-97c1-c58a6b1091dc" alt=""><figcaption></figcaption></figure></div>

Below the metrics, **Error Summary** shows where errors concentrate: the most frequent error types, and the traces and sessions with the most errors. It's the starting point for any investigation.

**Usage** shows what the agent relies on most: the documents it queried, the policies and routines it matched, and the tools it called. Use **Show lowest** to reverse the order and find content the agent rarely or never uses, which is usually a candidate for review.

**Recent Activity** lists the latest deployments and context updates, with who made each change and when. When behavior shifts suddenly, check here first for what changed.

### Config

**Config** shows the agent's full configuration, either as a **Form** or as **YAML**. The form groups it into sections:

* **Identity**: the **Agent Version** (the runtime version the agent runs on), **Agent Name**, **Description**, **Avatar**, and default **Language**.
* **System prompt**: the prompt that defines the agent's persona and ground rules. **Source** shows the prompt used, its labels, and the version pinned.
* **Context**: **Linked context** lists the policies, glossaries, routines and macros the agent uses. Select one to open it. **Relationships** sets **Priorities**, which decide which item wins when two apply at the same time: an item can take precedence over all routines or over specific items.
* **Extra rules**: the **Greeting** that opens the conversation, and preamble examples that shape the short messages the agent sends while it works.
* **Connections**: the models the agent uses and the external systems it connects to.
  * **Models**: **Provider / model** is the default for everything the customer reads, and **Fallback** lists the models tried, in order, when it fails. **Response** sets the model and **Reasoning** effort for the final reply. **Search query** sets the model that writes knowledge-base queries. **Evaluation models** cover the decisions the customer never sees: the default **Evaluation model**, the **Evaluation fallback** used when a decision keeps failing, and the models for **Tool calls**, **Policy matching** and **Routine navigation**. For the last two, a **Confidence threshold** sends uncertain decisions to the evaluation fallback for a second opinion. Any model left blank uses its group's default.
  * **Search**: the knowledge base the agent searches. It sets the **Database** and **Collection**, the **Fields** it queries (each with a **Type** and a **Field prompt** that turns the conversation into query text), how many results it returns (**Limit**), and how many recent customer messages it considers (**History limit**).
  * **Database**: the PostgreSQL database where the agent stores its sessions. When it's off, conversations are kept in memory and lost when the agent restarts.
  * **MCP connections**: the MCP servers attached to the agent, with the **Endpoint** used to reach each one (**Private**, within the project, or **Public**) and its credentials.
* **Advanced**: **Runtime** settings such as the **Log level**, which sets how much detail reaches the **Logs** tab; the **Deployment environment** recorded on each trace; the **Base URL** of the platform the agent loads its content from; and inbound webhooks. Leave fields blank to use the defaults.
* **Additional fields**: settings the agent version supports that the form doesn't place elsewhere, such as the **Variables** the agent expects its integration to set (for example, a ticket ID). They depend on the **Agent Version**.
* **Deployment**: the **Image**, the **Stack ID** of the stack the agent belongs to, the **Public URL** (with a **Hostname preview**), the schedule, the **Router API Key** the agent uses to call models through the Router, the **Agent API Key** your integrations use to call the agent, and the **Secrets** it loads. Changes here deploy directly to the live agent.

<div data-with-frame="true"><figure><img src="https://708770081-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1ICwJbq7EJdn5kBgXnQu%2Fuploads%2F2HjqMnFRbjPlY8dbugq2%2Fimage.png?alt=media&amp;token=754ee101-7144-457d-9082-ef8897b28603" alt=""><figcaption></figcaption></figure></div>

{% hint style="info" %}
The agent runs on exact versions of its context items. Publishing a new version of a policy, for example, changes nothing until that version is pinned to the agent and the agent is redeployed. See [Versioning & compatibility](https://docs.interactive.ai/agents/operations/versioning.md), [Priorities & entailments](https://docs.interactive.ai/agents/concepts/priorities.md), [Models](https://docs.interactive.ai/agents/concepts/models) and [Knowledge base & retrieval](https://docs.interactive.ai/agents/concepts/knowledge-base.md) in the Agents docs, and Router.
{% endhint %}

Values written as `${NAME}` in the YAML view, such as a database password, are read from the agent's [**Secrets**](https://docs.interactive.ai/infrastructure/secrets.md) instead of being stored in the configuration.

### Context

**Context** lets you browse the context items themselves, by type: **Policies**, **Routines**, **Macros**, **Prompts**, **Glossaries** and **Collections**. **Variables** are coming soon. Open a type to list its items with their number of **Versions**, **Tags**, last update, and the **Agents** that use each one, just like on the Project View.

<div data-with-frame="true"><figure><img src="https://708770081-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1ICwJbq7EJdn5kBgXnQu%2Fuploads%2F3vtVqcnfHRYGrd1Vug76%2Fpo-matching-agent-policies-tags.gif?alt=media&amp;token=00a6e5a7-7a33-4cbd-be13-cba5a1e43532" alt=""><figcaption></figcaption></figure></div>

Where **Config** shows which versions the agent pins, **Context** shows the items and their full history.

{% hint style="info" %}
Each context type has its own page: [Policies](https://docs.interactive.ai/context/policies.md), [Routines](https://docs.interactive.ai/context/routines.md), [Macros](https://docs.interactive.ai/context/macros.md), [Prompts](https://docs.interactive.ai/context/prompts.md), [Glossaries](https://docs.interactive.ai/context/knowledge.md) and [Collections](https://docs.interactive.ai/context/knowledge.md).&#x20;
{% endhint %}

### Tools

**Tools** lists the MCP servers connected to the agent under **Connected MCPs**, each with its status and number of tools. Select a server to see how many agents it's attached to (**Attached Agents**) and the description of each tool.

<div data-with-frame="true"><figure><img src="https://708770081-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1ICwJbq7EJdn5kBgXnQu%2Fuploads%2FoGI6G5VRW3aqKDaUk5a6%2Fimage.png?alt=media&amp;token=2c286cd5-83bb-4394-940a-00360b13e767" alt=""><figcaption></figcaption></figure></div>

Tool names carry the prefix the agent calls them by, for example `tools:send_email`. Use **Test** on a tool to run it on its own and check its response before a policy or routine relies on it.

{% hint style="info" %}
Tools are connected and managed in [Integrations](https://docs.interactive.ai/connectors/integrations.md).
{% endhint %}

### Traces & Sessions

**Traces** lists every turn the agent handled, with its **Input**, **Output**, **Latency**, **Tokens** and **Model Cost**. The **Metadata** of each trace records the agent configuration that was live for that turn, so you can link a change in behavior to the deploy that caused it.

**Sessions** lists the agent's conversations. Each session is one conversation and groups the traces of its turns. For each session you see its **Trace Status**, number of **Traces**, **Duration**, **Tokens** (input → output, with the total) and **Model Cost**. Filter by time range and by **Environment**.

<div data-with-frame="true"><figure><img src="https://708770081-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1ICwJbq7EJdn5kBgXnQu%2Fuploads%2FiygvijB6prI6IPFl6Gf0%2Fpo-matching-agent-traces.gif?alt=media&amp;token=aa8f037c-94ef-415c-b92f-26a4430d5267" alt=""><figcaption></figcaption></figure></div>

{% hint style="info" %}
Sessions and traces are covered in detail in [Observability.](https://docs.interactive.ai/governance/observability.md)&#x20;
{% endhint %}

### Logs

**Logs** shows the agent's runtime messages. Filter by level (**Errors**, **Warnings**, **Info**, **Debug**), date, search messages, and set how often the list refreshes. Check the logs when a trace doesn't explain a problem, such as a failed start or a model falling back to another one.

### Improvement

**Improvement** collects the review work on this agent's outputs:

* **Annotations**: the annotation queues for this agent, with their completed and pending items and the score configs reviewers fill in. Use **Process Queue** to review a queue's items.
* **Experiments**:  test runs that help you check whether an agent behaves correctlyand whether a change actually improves it.
* **Scores**: the scores recorded on this agent's traces, with their **Source** (such as **ANNOTATION**), **Name**, **Data Type** and **Value**.

<div data-with-frame="true"><figure><img src="https://708770081-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2F1ICwJbq7EJdn5kBgXnQu%2Fuploads%2FsK5mLTmdx5QEHBaMXNSI%2Fimage.png?alt=media&amp;token=ad873bc5-ed53-4444-bbbd-39c0a1c95541" alt=""><figcaption></figcaption></figure></div>

{% hint style="info" %}
See [Annotations](https://docs.interactive.ai/improvement/annotations.md), [Datasets](https://docs.interactive.ai/improvement/datasets.md) and [Scores](https://docs.interactive.ai/improvement/scores.md) for how each works across the project.
{% endhint %}

***

#### Ask the Copilot

| Example request                                                              | What the Copilot does                                                                |
| ---------------------------------------------------------------------------- | ------------------------------------------------------------------------------------ |
| "What changed between the last two revisions of the claims agent?"           | Compares the two revisions and summarizes the differences.                           |
| "Show me the errors the claims agent logged in the last hour"                | Fetches the agent's recent logs and lists the errors.                                |
| "Pin version 4 of the approval-threshold policy on the claims agent"         | Updates the agent's configuration with the new pin and redeploys it.                 |
| "Run the claims agent only on weekdays from 8:00 to 20:00, Madrid time"      | Sets an uptime schedule on the agent.                                                |
| "Replay the scenarios in the claims-replay dataset against the claims agent" | Re-runs the recorded conversations against the agent and reports which pass or fail. |


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.interactive.ai/interactive-agents/agent-view.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
