Switching To New AI Agent V5
Transitioning to AI Agent V5: Memory-Based Intelligence
Welcome to the next generation of automated customer support. AI Agent V5 moves away from static website scraping toward a dynamic, Memory-Based Learning system — now with full Sources support and a built-in Review tab to keep your AI accurate and consistent.
This article walks you through the updated interface and shows you how to set up your agent correctly
Watch the Video on the Best Practice here, or just read the steps down below:
1. Switching from V4 to V5
Navigate to your AI Agent settings.
The Toggle: In the top right corner, you will find the version toggle. Make sure V5 is selected and active.
Legacy Access: V4 is no longer in active use, but the interface remains accessible so you can retrieve your old instructions or escalation settings during the transition.
2. Key Differences: Why V5?
Memory-Based Learning: V5 learns from your actual conversations in the Inbox, not just static data sources.
Sources are back — and better: You can now feed the AI with structured knowledge from multiple source types (see Section 5).
Contradiction Detection: When two sources conflict, V5 flags the issue for your review instead of silently picking one answer (see Section 6).
Centralized Control: Instructions, Memories, Sources, and the Review queue are all managed from one place inside the AI agent settings.
3. Configuring Agent Instructions
The Agent Instructions field is the "brain" of your bot. It uses natural language — write it like you would brief a new support agent.
Consolidate your rules: Paste in your previous behavior rules, escalation logic, and tagging requirements.
Escalation Logic: Define specifically who handles what. Example: "If a customer asks about a return, escalate to @AgentName."
Auto-Save: The system saves your changes automatically.
4. Understanding Memories
The Memory List is where the AI stores facts it has learned from your conversations and sources.
Automatic creation: As the AI handles tickets, it builds memories from those interactions.
Manual control: You can edit existing memories, delete outdated ones, or add new memories manually to give the AI specific knowledge immediately.
Full transparency: Every memory shows which source it came from so you always know why the AI knows what it knows.
5. The Sources Tab
The Sources tab is where you give the AI structured knowledge to work with. This is separate from Memories — Sources are the raw material, Memories are what the AI extracts and uses from them.
You have four source types available:
Files
Upload internal documents directly to the AI. Supported formats: .pdf, .doc, .docx, .txt.
To upload: drag and drop your file into the upload area, or click "Select Files."
Note: If you are uploading a PDF, make sure the text in the document is selectable (not a scanned image), otherwise the AI cannot read it.
Website
Add URLs from your website, help content, or any relevant web pages. The AI reads and indexes the content from each link.
Help Center
Connect your existing Help Center to the AI. Articles are pulled in automatically and kept up to date as you publish or edit them.
AI Integrations
Connect third-party integrations such as Shopify so the AI can pull live product data, order information, and other shop-specific knowledge directly into conversations.
Retraining the Agent
After adding or updating sources, click Retrain Agent in the right panel to apply the changes. The panel shows you the current training status, total knowledge size, and any items queued for processing.
Storage limit: The total size of all sources combined is capped at 50 MB.
6. The Review Tab
The Review tab is new in V5 and solves one of the most common AI accuracy issues: contradicting information.
Why it exists
When two knowledge sources say different things (for example, a human agent wrote one answer and your website states something else), the AI previously had to pick one without any way to flag the conflict. This led to inconsistent responses.
With V5, conflicting memory pairs are automatically detected and moved into the Review queue. The AI will not use any disputed memory to answer customer questions until a human has resolved it.
How to review
Open the Review tab inside your AI agent settings.
You will see a list of flagged memory pairs. Each entry shows the two conflicting pieces of information and the sources they came from — click any source to view the full original content.
For each pair, choose which version is correct, or edit the memory manually.
Once resolved, the memory becomes active and the AI uses it again.
Clearing the backlog
If you are seeing a large number of items in your Review tab right now, those are conflicts that have been accumulating from previous months. Work through them when you can. The AI continues to operate normally in the meantime — it simply skips the disputed memories until they are resolved.
Once the backlog is cleared, expect around 3 to 4 new review pairs per week at most.
7. Tone and Behavior Settings
Below the instructions, you can fine-tune how the agent communicates:
Tone of Voice: Set the general style, for example Professional vs. Casual.
Emoji Usage: Define if and how often the agent uses emojis in replies.
Response Length: Control whether the agent gives quick, short answers or more detailed explanations.
8. Integrations (Shopify & More)
Existing integrations such as Shopify are fully compatible with V5. Make sure your integration shows as "Active" in the AI Integrations tab so the agent can pull product and order information directly into conversations.
Summary: What to do after reading this
Switch to V5 if you haven't already.
Add your knowledge sources under the Sources tab (Files, Website, Help Center, AI Integrations).
Click Retrain Agent after adding sources.
Open the Review tab and work through any flagged memory conflicts.
Check your Agent Instructions and make sure escalation logic is up to date.