Thai Open AI Project
Projects

Multi-Agent AI

PROTOTYPEInteractive laboratory · updated 2026-08-04

Thai Open AI Agent Lab

ห้องปฏิบัติการเอเจนต์

An observable environment where research agents run against real tasks with full logs of queue, latency, token usage and hand-off structure.

สภาพแวดล้อมที่สังเกตการทำงานของ AI Agent ได้แบบละเอียด

View ResearchDemoArchitecturePaperDatasetCodeAPI

Problem

Agent systems are opaque; failures are hard to attribute to planning, tooling or the model itself.

Research Question

What instrumentation is required to make multi-agent failures diagnosable?

Objectives

  • Expose per-agent queue, latency and token accounting
  • Record hand-off graphs for every task
  • Allow agents to be swapped without changing the interface

Methodology

  1. 01Instrumented runtime
  2. 02Task replay
  3. 03Failure taxonomy coding

Architecture

Task InputWeb console

Human-submitted research task

CoordinatorOrchestrator

Assigns and monitors agents

Agent Pool9 agent roles

Research, vision, data, reasoning

Trace StorePostgreSQL

Full execution traces

Results & Benchmark

Traced agent roles
9

Limitations

  • · Live agent activity shown on the public site is a simulation unless connected to a real server