06-05

Workflow

• Support defining component input parameter variable types.
• Start: Support importing input parameters in JSON format.
• LLM: Support importing input parameters in JSON format.
• HTTP Request: Support importing complete HTTP request configuration in cURL format.
• Added "Tool" component: Support adding self-built tools or GPTBots official tools as workflow nodes.
• Added "File Parser" component: Support parsing file-type variables into text output.
• Added "Text Processor" component: Support processing text (strings), such as concatenation, splitting, etc.
• Added "Variable Aggregator" component: Support aggregating different variables from multiple branches into one variable for easier use in subsequent processes.
• Added "Log" module: View workflow execution logs for troubleshooting.
• Added "Insights" module: Monitor workflow operation status through visual dashboards, including metrics such as runs, errors, success rates, etc.

Message Input Box

• Agent now supports clipboard image data, allowing direct pasting of images as attachments.
• FlowAgent and Workflow canvas now support Undo operations via Ctrl+Z shortcut, with a maximum of 50 steps that can be undone.

05-28

Chrome Extension

GPTBots Assistant is a Chrome browser extension built for GPTBots platform agents. After configuring the Agent's APIKey, it enables features like Human Handoff Notifications and Reply Suggestions.

The current version's Reply Suggestions feature supports the LINE platform, while Human Handoff Notifications are supported across all platforms in the integration module. For detailed information and extension download, please refer to GPTBots Assistant.

05-12

LLM Updates

• Added support for the Gemini-2.0-Flash-Image-Generation model, enabling LLMs to generate single or multiple images.
• Exposed advanced LLM parameter controls, allowing users to customize model-supported parameters such as Temperature, Top P, Top A, Top K, Presence Penalty, Frequency Penalty, Repetition Penalty, Min P, and more.

Agent Message Type Enhancements

GPTBots Agent now supports six types of input and output message formats: Text, Image, Audio, Video, Document, and File, making interactions with agents more flexible. The types of messages an agent can receive depend on the "Input-Attachment" configuration, while the types it can output depend on the capabilities of the underlying LLM.

Note: Supported formats for Image, Audio, Video, and Document messages may vary depending on whether "System Recognition" or "LLM Recognition" is selected.

FlowAgent Improvements

• Comprehensive optimization of the FlowAgent canvas creation experience, aligning interaction logic with workflow design.
• Added support for "Add Note" functionality, allowing users to annotate the canvas to help others understand the design and usage.
• Added the ability to insert new components in the middle of connection lines, enabling more flexible canvas design.
• Improved the component Prompt configuration UI for better clarity and ease of use.

Variable Reference

• Agents now support variable referencing in configurations (prompts, Flow components, welcome messages, etc.), including Global Variables, User Attributes, Browser Attributes, and Third-Party Platform Variables, enhancing agent flexibility and extensibility.
• Added a new global variable {{sys_user_msg_count}} to track the total number of messages exchanged between the user and the agent in the current conversation.
• FlowAgent now supports referencing Upstream Component variables in addition to the above, further improving flexibility and extensibility in FlowAgent configurations.

Zoho Sales IQ Integration

• Added integration support for the Zoho Sales IQ platform, allowing agents to be embedded in Zoho Sales IQ so users can interact with AI Agents directly through Zoho Sales IQ.
• Added support for Zoho Sales IQ in human handoff scenarios, supporting both Other to Zoho Sales IQ and Zoho Sales IQ to Zoho Sales IQ modes. After a user initiates a human service request, issues can be forwarded to Zoho Sales IQ customer service for resolution.

Knowledge Base QA Enhancements

• Knowledge Base QA now supports importing .xlsx and .xls files.
• Google Drive integration now allows users to select .csv, .xlsx, and .xls files for QA data import via the "Template Import" feature. Updates and management of spreadsheets in Google Drive are automatically synchronized to the AI Agent in GPTBots.

04-22

AI Agent knowledge source display

• Display knowledge documents and tool calls simultaneously
• Tag mode, incorporating knowledge source localization, allowing single-click tracking to the specific location of referenced knowledge in the original document
• Support for viewing the full text of referenced documents

Knowledge base slicing optimization

• Optimized knowledge base slicing display logic to enhance readability
• PDF, table, and docx documents now support locating the corresponding position in the original document and highlighting the corresponding section by selecting slices

Knowledge base upload - Tables

• A "custom row split" function has been added for table-type files, allowing users to split according to specified row numbers

Agent integration

• Support for the LINE platform, enabling the reception and recognition of user-sent messages, including text, images, audio, and documents

Log

• In Logs-Conversation, provides support for viewing user attribute information, including user ID, anonymous ID, email, and phone number

04-10

Newly developed Workflow module

Build intelligent workflows for AI applications through a simple drag-and-drop operation on a canvas. Your workflows can either be integrated into AI agents as tools, or serve as standalone APIs for seamless integration with any external program or system.

Feature optimization

• Upgraded all integration channels to V2 API, supporting conversation interruption mode.
• The model service protocol now supports OpenRouter services, enhancing LLM service stability.
• Voice message now supports 90s duration.
• Added support for automatic capture of message timestamps, enhancing LLM's temporal awareness during interactions.

03-28

• Supports customizing "Receive message type" in "Human handoff" for both Agent and FlowAgent.
• Supports setting "Receive message type" seperately for AI Agent interaction and Human handoff interaction.

03-26

• Variable usage: Supports the use of system predefined variables in agent configurations (prompts, Flow components, welcome messages, etc.), including global attributes (conversation ID, user ID, etc.), browser attributes (browser name, time zone, etc.), and third-party application attributes (WhatsApp, Telegram, and LiveChat client-related attributes, such as user ID, conversation ID, etc.). This helps maximize the agent's capabilities in different scenarios.

03-21

• When entering live chat for manual service through the Bubble Widget, users are allowed to send and receive text messages, audio messages, image messages, and document messages.
• The Bubble Widget supports uploading customized images to showcase the company's branding or background within the chat interface.

03-17

• Knowledge base/database supports global and knowledge base-level granularity permission configuration, improving data permission management efficiency.
• Optimized the display logic of knowledge base ID and data table ID.
• Inference models allow custom settings for reasoning depth and control over the display of the inference process, enhancing the interaction effects of the displayed content.
• The Webhook V2 version of the Dialogue - Send Message V2 API has been released, supporting audio messages, text messages, and detailed data on points and token consumption.

02-28

• The human service function module supports customizable prompt messages and multi-language configuration. The front end automatically adapts the display information based on the browser language.
• Optimized the bubble avatar display to a circular shape.
• Supports OpenAI project key authentication scheme.
• Released AI Agent Dialogue API V2, supporting text, audio, image, and document message types, including token and credit consumption details.
• Integration: Enables the agent to be integrated into Facebook Messenger and Instagram. Facebook and Instagram users can seamlessly interact with the agent through text, images, and voice.
• Supports OpenAI GPT 4.5 version & Anthropic Claude Sonnet 3.7 version.

02-17

• Support OpenAI Audio LLM, allowing users to input voice and directly reply with voice messages, eliminating the need for ASR and TTS.
• Text and voice conversations now support an "interruptible conversation mode," allowing users to send multiple messages in a short time while receiving a single response message, enhancing the user experience.

01-09

Agent Insights

Insights: H

• Real-time: Agent usage for the current day.
• Effectiveness: Metrics related to issue resolution effectiveness, manual intervention, user sentiment analysis, etc.
• Usage: Usage of credits and tokens.
• Users: Metrics such as user count, new users, active users, etc.
• Behavior: Metrics related to conversations, interactions, etc.
• Health: Metrics such as response speed, errors, and requests per minute.