Changelog

Your custom API tool integrations are now protected by two new security layers, and you can test them directly from the dashboard before going live.

1. API Response Security Custom API functions now validate every response before it reaches the AI, ensuring your chatbot never processes unexpected or dangerous content.

• Content-Type Blocking: Only text and JSON responses are accepted. Binary files (PDFs, images, executables) are automatically blocked before any data is read.
• 50 KB Response Cap: Responses are streamed in chunks and capped at 50 KB to prevent abuse. Oversized responses are truncated gracefully so the AI still gets useful data.
• Accept Header: All outgoing API requests now include an `Accept` header signaling that only text/JSON is expected.

2. Test API — Try Before You Deploy A new Test API panel inside the Custom Function editor lets you validate your API configuration instantly — without saving or deploying.

The Test API panel with auto-filled arguments and live results.

• Smart Auto-Fill: Test arguments are auto-generated from your query parameters and JSON schema. Required fields get type-appropriate placeholders, and `e.g.` hints from descriptions are extracted automatically.
• Live Results: See success/failure status, the full response body, and exact response time in milliseconds.
• Security in Action: Test results show the same security guards your live chatbot uses — blocked content types and truncated responses are clearly surfaced.

Security in action — a blocked binary download response.

We’ve introduced Ticket Only mode, a new way to manage human support requests that gives your team total flexibility. Instead of forcing a live chat connection, you can now configure your AI agents to capture inquiries as structured tickets that your team can solve whenever they are ready.

1. Professional Lead & Context Capture When a user requests human help, the AI gracefully transitions into a ticketing flow. It intelligently asks for the user's email address and provides them with a unique Ticket ID (e.g., TKT-0001). This ensures your team has everything they need—including the full chat transcript—to resolve the issue via email later.

• Automated Lead Capture: When human agents are unavailable, the AI intelligently switches to a 'Ticket Only' flow, asking for the user's email to follow up.
• Instant Ticket Creation: Once an email is provided, the system automatically converts the chat into a ticket with a unique ID (e.g., TKT-0001).

The widget seamlessly captures user contact info and provides a Ticket ID for follow-up.

2. Smart Availability & Scheduling You no longer have to worry about "missed chats" when your team is away.

• Always Available: Perfect for teams who prefer solving issues instantly.
• Scheduled Schedule: Define your business hours and timezone; the AI will automatically switch to Ticket mode when you're offline.
• Ticket Only: A simple toggle to instantly move all human handoffs to the ticketing system.

Enable "Ticket Only" mode

3. Reduced Pressure, Better Support This update is designed to bridge the gap between AI automation and human expertise. By moving to an asynchronous model, your team can manage their support queue at their own pace without the pressure of being "always online" for live chats, while your customers still receive a professional and guaranteed path to resolution.

Every offline request is automatically converted into a structured ticket with a unique ID, ensuring your team has the full context needed for follow-up.

We have significantly improved the user experience of the chat widget by making the AI's "thought process" transparent and delivering responses word-by-word. 1. Granular Processing Status Instead of a generic typing indicator, users now see exactly what the AI is doing in real-time. This transparency reduces perceived latency and provides immediate feedback after a message is sent.

• Rewriting Query: Shown when the AI is analyzing conversation history to understand the user's intent better.
• Fetching Content: Shown when the system is actively searching the knowledge base for relevant documents.
• Generating Answer: Shown just as the LLM begins to formulate the final response.

GIF showing the new widget with real-time streaming and intelligent status.

2. Word-by-Word Streaming Responses are now delivered word-by-word as they are generated. This allows users to start reading the answer immediately, making the widget feel significantly faster and more interactive compared to waiting for the full block of text.

3. Intelligent Auto-Scrolling The chat window now intelligently follows the AI's output to ensure the user never loses their place.

• Adaptive Scrolling: We've implemented high-performance "instant" scrolling during streaming to ensure the latest text is always in view without animation lag.
• Smooth Transitions: Standard messages and human agent typing indicators still use polished smooth-scroll animations.

4. Cross-Platform Style Fixes We've resolved a critical CSS inheritance issue that affected the widget when embedded in dark-themed Next.js or Tailwind projects.

Keeping your AI's knowledge base up-to-date just got effortless. We've introduced a powerful Web Crawling engine that can ingest content from single URLs or entire sitemaps, complete with automated scheduling.

Manage all your web sources and their sync status in one place.

• Sitemap & URL Support: Point us to a single page or a full sitemap XML to instantly ingest hundreds of pages.
• JavaScript Rendering: Our crawler handles dynamic, JS-heavy websites to capture content that simple scrapers miss.
• Granular Control: Hand-pick specific pages to crawl or set up exclusion patterns to skip irrelevant sections (like /admin or /login).

Select only the pages you want

• Auto-Sync Schedules: Set your sources to refresh Daily, Weekly, or Monthly so your AI always answers with the latest information.

Easily configure sync frequency when adding a new source.

This update ensures your chatbot evolves alongside your website without any manual intervention.

Managing multiple AI agents just got significantly easier. We've added three highly requested features to give you total control over your agent configurations.

New management tools available in the Agent settings menu.

• One-Click Duplicate: Instantly clone an existing agent's settings, knowledge base, and personality to test new variations.
• Agent Export: Download your agent's configuration as a JSON file for backup or version control.
• Seamless Import: Move agents between different organizations or accounts by simply uploading an export file.

Whether you are scaling your support team or moving agents between environments, these tools will save you hours of manual setup.

We are excited to announce our official Freshchat integration! You can now connect your Deflect AI agents directly to your Freshchat account to provide seamless, automated support across all your customer channels.

Easily connect your Freshchat account with a single API key.

• Intelligent handoff: AI handles the initial query and passes to a human agent in Freshchat when needed.
• Full conversation history preserved for human agents.
• Intelligent handsoff can be done within Deflect itself. This is just an addon integration for customers willing to stick to Freshchats

Turn on freshchat integration inside human handsoff of specific agent you want it for. Select the Freshchat channel where you want the chats to redirect

This integration ensures that your team stays in their preferred environment while leveraging the power of Deflect AI to reduce ticket volume.