Skip to main content

Overview

Chat is Wegent's core interaction feature that allows you to have real-time conversations with AI agents to complete various tasks.

Chat Mode Demo

πŸ“‹ Core Concepts​

Chat Tasks​

A Chat Task is the basic unit of the Chat feature. Every interaction with AI creates a task:

  • Single-turn: Simple one-time Q&A
  • Multi-turn: Continuous contextual conversation
  • Code Task: Programming tasks related to code repositories

Chat Modes​

Wegent provides multiple conversation modes for different scenarios:

ModeDescriptionUse Case
Smart Follow-upAI asks questions to clarify requirementsComplex tasks with unclear requirements
AI Cross-ValidationAI suggests modificationsCode review, document optimization
Direct ExecutionAI executes tasks immediatelySimple tasks with clear requirements

🎯 Main Features​

1. Task Management​

Centralized management of all chat tasks:

  • View conversation history list
  • Filter by status (in progress, completed, failed)
  • Search and archive tasks
  • Export conversation records

2. Multi-Agent Collaboration​

Support for various collaboration methods:

  • Single Agent: One-on-one conversation with one agent
  • Group Chat: Multiple agents participating in discussions
  • Team Mode: Agents collaborate with assigned roles

3. Code Tasks​

Enhanced features designed for programming:

  • Automatic code repository cloning
  • Agents read and analyze code
  • Execute code modifications and commits
  • Real-time execution monitoring

4. Conversation Control​

Flexible conversation control options:

  • Pause/resume conversation
  • Regenerate responses
  • Modify context
  • Switch agents

5. Runtime Model Switching​

After a task has started, you can adjust the model used for later responses in the current task. The change only affects the current task. It does not modify the agent, bot, or model defaults, and it does not affect other tasks or newly created conversations.

Wework supports switching between official GPT/Codex models and third-party models in the same conversation. After confirmation, the new model applies to the next normal message. A response already in progress still finishes with the previous model.

Encrypted reasoning and compaction state is provider-specific and cannot be reused directly. At the switch boundary, the executor removes the previous provider's encrypted_content and previous_response_id while retaining portable messages, tool results, and local context. This prevents the target model from returning invalid_encrypted_content.

πŸ“– Documentation Navigation​

DocumentDescription
Managing TasksCreating and managing chat tasks
Smart Follow-up ModeUsing smart follow-up mode to clarify requirements
AI Cross-ValidationUsing AI cross-validation to optimize results

πŸš€ Quick Start​

Create Your First Conversation​

  1. Navigate to the Chat page
  2. Click New Conversation
  3. Select an agent or agent team
  4. Enter your question or task description
  5. Wait for AI response

Create a Code Task​

  1. Click New Task
  2. Select a programming-type agent
  3. Select a code repository
  4. Describe the task to complete
  5. AI will automatically clone the repository and start working

πŸ’‘ Use Cases​

Programming Assistance​

  • Code Review: Have AI check code quality and potential issues
  • Bug Fixing: Describe the problem, let AI locate and fix it
  • Feature Development: Describe requirements, let AI implement new features
  • Code Explanation: Have AI explain complex code logic

Content Creation​

  • Documentation Writing: Have AI help write technical documents
  • Content Polishing: Optimize existing content expression
  • Translation Services: Multi-language content translation

Knowledge Q&A​

  • Technical Consulting: Ask technical questions and best practices
  • Concept Explanation: Understand complex technical concepts
  • Solution Design: Discuss system architecture and design solutions

πŸ”— Related Resources​