DEV Community

Jeffrey.Feillp
Jeffrey.Feillp

Posted on

Tian AI Autonomous Agents: Task Scheduling with LLM

#python #ai #agents

Tian AI Agent System: Autonomous Task Scheduling with LLM

One of Tian AI's most powerful features is its autonomous agent system — an LLM-driven task scheduler that plans, executes, and adapts tasks without human intervention.

The Problem

Traditional AI assistants can only respond to queries. Tian AI's agent system proactively executes multi-step tasks, making it useful as a real productivity tool.

Architecture 

User Request
     ↓
  LLM Parser → Intent Classification
     ↓
  TaskQueue → Dependency Resolution
     ↓
  Safety Whitelist → Permission Check
     ↓
  Task Execution (tools / Python / shell)
     ↓
  Self-Reflection Loop
     ↓
  Result → Summary
Enter fullscreen mode Exit fullscreen mode

Key Components

LLM Parser

Uses Qwen2.5-1.5B to understand natural language requests and decompose them into actionable tasks. Intent categories include plan, execute, search, ask, and greeting.

TaskQueue with Dependency Resolution

Tasks can have dependencies. The queue uses topological sorting to determine execution order. Example:

Task A: "Check disk space" → no deps
Task B: "Backup database" → depends on A
Task C: "Send report" → depends on A and B
Enter fullscreen mode Exit fullscreen mode

Safety Whitelist

Critical for autonomous execution. The whitelist defines:

  • Allowed directories (project root only)
  • Allowed shell commands (no rm -rf, no sudo)
  • Read-only by default; write requires explicit permission
  • Network access to dev.to / huggingface.co only

Self-Evaluation Loop

After each task execution, the system evaluates:

  1. Did the task complete successfully?
  2. Is the output consistent with expectations?
  3. Should alternative approaches be tried?

Greeting Shortcut

For efficiency, common greetings bypass the LLM entirely:

GREETING_PATTERNS = [
    r"(?i)^(hi|hello|hey|你好)\b.*$",
    r"(?i)^(how are you|你|好|在吗)\b.*$",
]
Enter fullscreen mode Exit fullscreen mode

Example: Autonomous Workflow

User says: "Generate a promotional video and post it to Dev.to"

  1. LLM parses → 2 subtasks (generate video, publish article)
  2. TaskQueue resolves dependency: video must be generated before posting
  3. Safety whitelist allows: video generation ✓, but posting blocked (no write permission for dev.to)
  4. System reports: "I can generate the video autonomously, but posting requires your confirmation"

This safety-conscious design is what makes Tian AI trustworthy for autonomous operation.

Published on 2026-04-25 21:19 UTC by Tian AI Dev Team

Top comments (0)