Skip to content

Agent System Overview

Understanding the AI agent system in tg-note.

What are Agents?

Agents are AI-powered systems that process your messages and transform them into structured knowledge base entries. They analyze content, categorize it, extract key information, and generate well-formatted Markdown files.

Available Agents

tg-note supports three types of agents, each with different capabilities and use cases.

Comparison Table

Feature Qwen Code CLI Autonomous Stub
AI Processing ✅ Advanced ✅ Advanced ❌ Basic
Auto Planning ✅ Yes ✅ Yes ❌ No
Web Search ✅ Built-in ✅ Built-in ❌ No
Git Operations ✅ Built-in ✅ Built-in ❌ No
Function Calling ✅ Yes ✅ Yes ❌ No
External Dependencies Node.js OpenAI lib None
Free Tier 2000/day Depends on API Free
Best For Production Custom APIs Testing/MVP

The most powerful agent using the official Qwen Code CLI tool.

Features

  • ✅ Advanced AI Processing - Qwen3-Coder models
  • ✅ Automatic Planning - Creates and executes TODO plans
  • ✅ Built-in Tools - Web search, Git, GitHub, Shell
  • ✅ Vision Support - Can analyze images
  • ✅ Free Tier - 2000 requests/day, 60 req/min

Installation

# Install Node.js 20+
npm install -g @qwen-code/qwen-code@latest

# Authenticate
qwen

# Configure
AGENT_TYPE: "qwen_code_cli"

When to Use

  • ✅ Production deployments
  • ✅ Need best quality output
  • ✅ Want automatic planning
  • ✅ Can install Node.js

Full Documentation →

2. Autonomous Agent

Python-based agent with OpenAI-compatible API support.

Features

  • ✅ OpenAI-Compatible - Works with OpenAI, Qwen, and other compatible APIs
  • ✅ Autonomous Planning - Self-planning and decision-making
  • ✅ Function Calling - Native function calling support
  • ✅ Built-in Tools - Web search, Git, GitHub, file management
  • ✅ Flexible - Custom LLM connectors

Installation

# OpenAI library included in requirements
poetry install

# Configure
AGENT_TYPE: "autonomous"
AGENT_MODEL: "gpt-3.5-turbo"  # or any compatible model

Configuration

Environment Variables: 

OPENAI_API_KEY=sk-...
OPENAI_BASE_URL=https://api.openai.com/v1  # Optional

YAML Settings: 

AGENT_TYPE: "autonomous"
AGENT_MODEL: "gpt-3.5-turbo"
AGENT_ENABLE_WEB_SEARCH: true
AGENT_ENABLE_GIT: true
AGENT_ENABLE_FILE_MANAGEMENT: true

When to Use

  • ✅ OpenAI or compatible API available
  • ✅ Need autonomous planning
  • ✅ Custom LLM provider (Qwen, Azure, etc.)
  • ✅ Python-only environment
  • ❌ Don't want to use Node.js

Full Documentation →

3. Stub Agent

Simple testing agent without AI.

Features

  • ✅ Fast - No API calls
  • ✅ Simple - Basic categorization
  • ✅ No Dependencies - Works out of the box
  • ❌ Limited - No AI analysis

Configuration

AGENT_TYPE: "stub"

When to Use

  • ✅ Testing and development
  • ✅ MVP/prototype
  • ✅ No API keys available
  • ❌ Production use

Full Documentation →

How Agents Work

Processing Pipeline

graph LR
    A[Message] --> B[Agent]
    B --> C{Analyze}
    C --> D[Categorize]
    D --> E[Extract Info]
    E --> F[Generate Markdown]
    F --> G[Save to KB]

    style B fill:#fff3e0
    style F fill:#e8f5e9

Agent Workflow

  1. Receive Content
  2. Text messages
  3. URLs
  4. Forwarded posts

  5. Media

  6. Analyze

  7. Understand topic
  8. Extract key points

  9. Identify category

  10. Process

  11. Search for context (if enabled)
  12. Gather additional info

  13. Structure content

  14. Generate

  15. Create Markdown file
  16. Add metadata

  17. Format properly

  18. Save

  19. Write to KB
  20. Commit to Git
  21. Notify user

Choosing an Agent

Decision Tree

graph TD
    A[Choose Agent] --> B{Production or Testing?}
    B -->|Production| C{Can install Node.js?}
    B -->|Testing| G[Stub Agent]
    C -->|Yes| D[Qwen Code CLI ⭐]
    C -->|No| E{Have OpenAI-compatible API?}
    E -->|Yes| F[Autonomous Agent]
    E -->|No| G

    style D fill:#c8e6c9
    style F fill:#b3e5fc
    style G fill:#ffccbc

Recommendations

For Production

→ Qwen Code CLI

  • Best quality results
  • Automatic planning
  • Built-in tools
  • Free tier available

For Testing

→ Stub Agent

  • Quick setup
  • No dependencies
  • Fast iteration

Agent Configuration

Global Settings

Set in config.yaml:

# Agent Selection
AGENT_TYPE: "qwen_code_cli"

# Common Settings
AGENT_MODEL: "qwen-max"
AGENT_TIMEOUT: 300

# Tool Permissions
AGENT_ENABLE_WEB_SEARCH: true
AGENT_ENABLE_GIT: true
AGENT_ENABLE_GITHUB: true
AGENT_ENABLE_SHELL: false

Per-User Settings

Users can override via Telegram:

/agentsettings

Agent Capabilities

Content Analysis

All agents can: - Extract text content - Identify topics - Generate titles - Create metadata

Advanced agents (Qwen) add: - Deep semantic understanding - Context enrichment - Smart categorization - Quality summaries

Tool Usage

Tool Qwen CLI Autonomous Stub
Web Search ✅ ✅ ❌
Git Ops ✅ ✅ ❌
GitHub API ✅ ✅ ❌
Shell ✅ ✅ ❌
File Management ✅ ✅ ❌
Vision ✅ ✅* ❌

*Vision support depends on the LLM model used

Performance

Typical Processing Time

Agent Short Text Medium Text Long Text
Qwen CLI 5-15s 15-45s 45-120s
Autonomous 5-20s 15-60s 60-180s
Stub <1s <1s <1s

Factors Affecting Speed

  • Content length
  • URL complexity
  • Web search usage
  • API response time
  • Network latency

See Also