The AI Agent is TxtCart's built-in AI that handles customer conversations over SMS, answering questions, resolving support requests, and driving revenue around the clock. You can fully customize how it behaves, what it knows, and what it says.
​

To access it, click "AI Agent" in the left side menu. From there, you'll find six tabs across the top: Persona, Skills, Knowledge Base, Verified Answers, Guidelines, and AI Playground.

This is where you define your agent's identity, voice, and style.

The Brand Summary field is where you describe your company's story and what your brand represents, in a couple of sentences. This is the foundation the AI agent builds its voice on. It has a 1,200-character limit.

Name: The display name your agent uses when interacting with customers.

Role: A dropdown that sets the agent's job title or function, like "Sales Associate."

Gender: Sets the agent's gender identity, which can influence how it refers to itself in conversations.

Age: A numeric field for the agent's perceived age, which can subtly shape the tone and relatability of responses.

Four sliders let you dial in how the agent should sound. Each sits on a spectrum between two opposing styles.

​Casual vs. Formal: Controls how relaxed or professional the language is.

Gentle vs. Bold: Determines how soft or assertive the agent's delivery is.

Direct vs. Empathetic: Balances getting to the point vs. acknowledging the customer's feelings.

Serious vs. Playful: Controls the overall mood and humor level of responses.

Words and Phrases: Add signature expressions or brand-specific language the agent should incorporate.

Sign-offs: Set how the agent closes out conversations.

Emojis: Add specific emojis you want the agent to use regularly.

Emoji Usage: A slider ranging from No Emojis to Many Emojis that controls how frequently emojis appear in responses.

Words and Phrases: Add competitor names, sensitive terms, or any language you want kept out of responses entirely.

Emojis: Block specific emojis from appearing in agent replies.

Skills are the specific actions your agent can take during a conversation. Each skill can be toggled on or off. Some skills also have a settings icon for additional configuration.

​Escalate to Human (Always on): Routes the conversation to your team when a subscriber requests it or a potential concern is detected. This skill cannot be turned off.

Search Products: Searches your Shopify catalog to answer product questions in real time.

Check Inventory: Provides real-time answers about product availability, including variant-level details.

Track Order / Shipment: Checks order and shipping status, then delivers updates to the customer via SMS.

Upsell / Cross-Sell: Recommends complementary products based on the subscriber's interests, using your bundles and collections. This skill has additional settings you can configure via the gear icon.

Cancel Order: Allows customers to cancel unfulfilled orders directly through SMS. Toggled off by default.

Initiate Return / Exchange: Starts the return or exchange process for eligible orders. Toggled off by default.

This is where you manage all the content your AI agent draws from when answering customer questions. The tab has two subtabs: Sources and Facts.

Sources are the documents, pages, and data feeds that power your agent's knowledge. Each source shows its name, type, tag, word count, sync frequency, and current status. You can search sources by name and filter by source type or status using the dropdowns at the top.

Each source entry shows:

The right sidebar shows a Sync Overview with a count of all your sources broken down by status, and a Source Types breakdown showing how many sources you have of each type.

Click "+ Add Source" in the top right, or use one of the Quick Actions in the right sidebar. You'll see five options:

Sources set to auto-sync, like Shopify Policies, will update daily without any manual action needed.

Facts are authoritative statements or information your AI agent always treats as ground truth. Use them to override ambiguous or conflicting information from your connected sources.
​
Each fact entry shows the statement, a category tag, and when it was last updated.
​
You can edit or delete any fact using the icons on the right.

Click "+ Add Fact" in the top right. A modal will appear with two fields:
​

Good candidates for facts include your support email, shipping carrier and timeframes, return policy exceptions, or anything that needs to be stated consistently regardless of what your other sources say.

Verified Answers are pre-approved question and answer pairs. When a customer's message matches a verified answer, the agent responds with it directly, bypassing the AI entirely. This ensures speed and consistency for questions you always want answered a specific way.
​

You can search existing verified answers using the search bar at the top and add new ones with the "+ Add Answer" button.
​

Each entry shows the question, the pre-written answer, a match type tag, a usage count, and the last updated timestamp. Entries with Skip AI Formatting enabled are delivered exactly as written, with no rephrasing by the agent.

Click "+ Add Answer" to open the form. You'll see the following fields:

Answer Variants (optional): Alternative phrasings of the answer so responses feel more natural across different conversations.

Skip AI Formatting: When toggled on, the answer is returned exactly as written, with no rephrasing by the AI.

Match Confidence: A dropdown that controls how closely a customer's message needs to match before this answer triggers. There are three options:

Good candidates for verified answers include payment method questions, tax or fee policies, shipping timeframes, and any question where you want the response to be word-for-word consistent every time.

Guidelines are behavioural rules your AI agent always follows. Use this tab to set hard rules around escalation policies, prohibited topics, required disclaimers, or anything the agent should consistently do or avoid regardless of the conversation.

Click "+ Add Guideline" to open the form. You'll see two fields:

Good candidates for guidelines include topics the agent should never comment on, situations that always require a human handoff, required language for sensitive topics, and any behaviour that needs to be consistent no matter what the customer says.

The AI Playground is where you test your agent before it goes live. You can send it questions and see exactly how it would respond to a real customer, based on your current Persona, Knowledge Base, Verified Answers, Guidelines, and Skills configuration.
​

The playground includes a few example questions to get you started: "Return policy?", "Do you accept PayPal?", "Best discount?", and "Ship times?" You can click any of these or type your own question in the input field at the bottom.
​

If responses aren't coming out the way you expect, there's a direct link at the bottom to Update Knowledge Base without leaving the playground.
​

To simulate another customer conversation from scratch, click "Start a new chat" in the bottom right. This keeps each test session clean and separate.

Click "Set Test Scenario" in the top right to simulate how the agent would behave depending on how a conversation was started. You can select from three options:
​

This is useful for making sure your agent responds appropriately depending on the context a customer is coming from, since behaviour may differ between a post-campaign reply and an organic inbound message.
​

Still have questions?
Please feel free to reach out to our wonderful Support team at support@txtcart.ai or via Live Chat.
​