> 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/build/agent-configuration.md).

# Agent Configuration

Agent Configuration is where you define **how your Interactive Agents behave**. Every agent starts as a configuration that captures its identity, its system prompt, the context it draws from, optional conversational rules, and the external services it talks to. Configurations are versioned, so every change creates a new revision you can compare, roll back to, or promote to production independently of the running deployment.

Navigate to **Build → Agent Configuration** to see all agent configurations in your project. The listing shows each agent's name, status, and last-updated time. If a draft shares a name with a deployed agent, the draft row replaces the deployed row until you publish.

<div data-with-frame="true"><figure><img src="/files/4c1Ogl5S0nNzZadTDAq2" alt=""><figcaption></figcaption></figure></div>

### Creating an Interactive Agent

Click **Create agent** in the Agent Configuration listing to open the creation wizard. The wizard walks through six steps. At the bottom of every step you can Cancel, Save draft, go Back, or Continue to the next step.

{% stepper %}
{% step %}

### Step 1: Identity&#x20;

Identity establishes how the agent identifies itself.

* **Agent name** is the identifier for the agent. It accepts letters, digits, hyphens, and underscores. The name is set at creation time and cannot be changed afterwards.
* **Language** controls what language the agent responds in. Choose a specific language (English, French, Spanish, German, among others) or select Match user's language to have the agent mirror the language of whoever is talking to it. The language can be changed later.
  {% endstep %}

{% step %}

### Step 2: Description

The Description is the system prompt that tells the agent who it is. It defines the agent's role, tone, and overall behavior. You have two options for providing it:

* **Write new** lets you type the description directly. The platform saves the text as a new prompt in your project's Context library so it can be versioned and reused. You can optionally set a Prompt name for this new prompt; leave it blank and the platform auto-generates one based on the agent name.&#x20;
* **Pick existing** lets you select a prompt that already exists in your project's Context library. A searchable list displays all available prompts with their labels and version numbers. The runtime binds to the exact version you pick here. If no prompts exist yet, the view shows a message directing you to switch to Write new or create a prompt in the Context page first.

{% hint style="info" %}
When editing an existing agent's description, any change to the linked prompt body creates a new version of that prompt automatically.
{% endhint %}
{% endstep %}

{% step %}

### Step 3: Context

Context attaches the reusable behavioral building blocks from your project's catalog. The agent fetches the live version of each item at runtime. This section is divided into two areas.

**Linked context** is where you pull in policies, glossaries, and routines from the project catalog. Each type has its own collapsible panel with a search field:

* **Policies** are behavior rules defined as condition-action pairs that the agent must respect across every conversation. For example, a data-protection policy might require the agent to never share another employee's personal information.
* **Glossaries** are collections of domain-specific terms the agent should recognize and use correctly. For example, an HR glossary might define terms like PTO, Flex Friday, or OnePlan so the agent uses them consistently.
* **Routines** are step-by-step procedures the agent follows for specific kinds of requests. For example, a time-off routine might walk an employee through the steps of requesting paid leave.

{% hint style="info" %}
Each attached item shows its name, version number, and a link icon that opens the item in the Context editor. You can remove any item with the X button next to it.
{% endhint %}

**Routine relationships** appears below the linked context and governs how routines interact with each other. It contains two controls:

* **Routine priorities** defines explicit precedence rules between routines. Only routines already attached to this agent can be referenced.&#x20;
* **Reevaluate routines after** specifies tools whose execution should trigger the agent to reconsider which routine applies. This is useful when a routine's activation depends on the result of a tool call. For example, after an authentication tool confirms the user's identity, the agent might need to switch from a general-inquiry routine to an account-management routine. The dropdown lists tools called by the routines attached above.
  {% endstep %}

{% step %}

### Step 4: Extra Rules

Extra Rules is an optional section that controls the agent's greeting and conversational style on the first turn. The entire section can be toggled on or off with the Enable extra rules switch. When disabled, the rules block is omitted from the configuration entirely and the section shows only the toggle.

When enabled, all three fields are required together. Partial rule sets are rejected.

* **Opening greeting** is the message the agent sends verbatim at the start of every new conversation. For example: "Hi, I'm Alex from the HR team. How can I help you today?"
* **Thinking phrases** are short acknowledgment phrases the agent uses while tools are running. At least one is required. You can add more with the + Add button. Examples: "Let me check.", "One moment.", "On it!"
* **Language instruction** is a natural-language sentence injected into the engine before each preamble, describing how the agent should phrase its responses. For example: "Always respond in the same language the customer is using."
  {% endstep %}

{% step %}

### Step 5: Connections

Connections is where you wire up external services and optionally override the default model. The section is divided into two areas.

**Model** lets you override the LLM provider and model the agent uses. A searchable dropdown lists every model available in the project's router catalog. Leave it on Default to use the agent's built-in choice.

**Endpoints** groups the external services the agent can reach at runtime:

* **Knowledge base** is an optional retriever endpoint with a prompt that selects what to surface. Toggle it on to reveal fields for a Retriever prompt id, a Hostname, a Port, and a Path.
* **Session stores** are external stores for conversation and session state. The counter shows how many are configured. Click + Add to attach one.
* **MCPs** are Model Context Protocol servers that expose tools to the agent. The counter shows how many are attached. Click Manage to pick from the connections registered on the project's Integrations page. If no MCP connections exist yet, a link directs you to the Integrations page to add one.
* **Integration webhook** is an optional webhook that receives lifecycle events from the agent. Toggle it on to reveal fields for a Hostname, a Port, and a Path.
  {% endstep %}

{% step %}

### Step 6: Review

The review screen shows five summary cards (Identity, Description, Context, Extra Rules, Connections) alongside a live preview of the Generated configuration in YAML. Each card has an Edit button that jumps back to the corresponding step. From here you have two options:

* **Save draft** stores the configuration without deploying it. The agent appears in the Agent Configuration listing as a draft and can be edited further before going live.
* **Save draft and deploy** stores the configuration and opens the deployment wizard under Deploy → Interactive Agent, where you configure the runtime details and bring the agent to production. The deploy step validates everything before anything actually deploys.
  {% endstep %}
  {% endstepper %}

### Versions and Editing Formats

Every time you save a change to an agent configuration, InteractiveAI creates a new version. Previous versions are preserved and accessible from the Versions panel on the left side of the detail view. Each version shows its number, the agent name, the date it was created, and its current status.

<div data-with-frame="true"><figure><img src="/files/sEh8RQr2eTojzC10OPpA" alt=""><figcaption></figcaption></figure></div>

You can search versions by number and click any version to view its full configuration. The detail view always shows which version you are looking at in the header (for example, "#9 demo-agent").

When you open a deployed version for editing, a banner at the top warns: "Editing the live agent. Save draft to keep working on a pending version without affecting the live agent. Deploy applies the changes to the live configuration immediately."

<div data-with-frame="true"><figure><img src="/files/z704qJJ3IWi3EytjPMBm" alt=""><figcaption></figcaption></figure></div>

The detail view offers two editing formats, toggled with the Form / YAML tabs at the top. Form mode displays the configuration as five collapsible sections (Identity, Description, Context, Extra Rules, Connections) with fields laid out visually, and is the default. YAML mode displays the raw `agent_config` representation, which you can edit directly; the platform validates the schema before accepting changes. Both modes always reflect the same underlying data, and changes in one are visible immediately when switching to the other.

### Editing and Redeploying

To edit an existing agent, open it from the Agent Configuration listing. You can update fields in Form mode or switch to YAML mode for a direct edit. At the bottom of the editor, three actions are available:

* **Cancel** discards unsaved changes and returns to the listing.
* **Save draft** creates a new draft version without affecting the live deployment. Use this when you want to iterate on changes before promoting them.
* **Deploy configuration** saves the configuration and applies it to the live agent immediately.


---

# 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 dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.interactive.ai/build/agent-configuration.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
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.