Upgrade the "Chatbot" into
Personal AI Operating System
AG2O isn\'t an AI in a window — it\'s a Personal Agent system that divides tasks, accumulates experience, and runs persistently. Move your AI from "answering questions" to "taking on work, assigning tasks, and building expertise".
$ Starting AG2O Daemon System...
[OK] Personal Agent [UID-001] Listening on Feishu/Slack
[ACTIVE] Task Agent [Researcher] processing 3 jobs
[FORKED] 4 Sub-agents running parallel research
// Workspace: a2_data/workspaces/personal
Intent recognized: User needs to sync next week supply chain report to Task Agent [Operations]
Always Online
Runs as a daemon, independent of chat windows, continuously monitoring environment and triggers.
Layered Collaboration
Clear Personal → Task → Forked three-layer architecture, each with its role, efficient concurrency.
Continuous Accumulation
Independent workspace with persistent sessions and metadata — a true professional persona that learns.
System Architecture
Three-Layer Agent System
More Than an Assistant — A Team
AG2O uses a persistent daemon plus three-layer Agent structure. Top layer understands the user, middle layer specializes in domains, bottom layer handles complex parallel execution.
Personal Agent
Front-end Understanding & Routing
Faces the user, responsible for receiving messages, understanding intent, managing relationships, and routing tasks.
Task Agent
Back-end Professional Execution
A long-resident specialized Agent for code, data, research and other domain tasks, continuously accumulating context.
Forked Agent
Bottom-Layer Parallel Breakdown
Execution units temporarily forked from Task Agent for handling parallel subtasks.
Personal Agent
Intent / Relationships / Task Routing
Task Agent A
Long-resident / Domain Expert
Task Agent B
Workspace / Memory Accumulation
Hardcore Runtime Capabilities
Persistent Daemon Service
Continuously listens to channel events and exposes REST API after startup. AG2O is a persistent personal AI service running in the background — not a temporary assistant that starts and stops on demand.
Three-Layer Agent Architecture
Personal Agent understands intent, Task Agent executes domain tasks long-term, and Forked Agent breaks down parallel subtasks — clear division of responsibilities.
Long-Term Workspace & Memory
Each Agent has its own workspace and persistent session records. It's a professional persona with long-term memory — not an executor that forgets everything when a task ends.
Real Input Processing Pipeline
Integrates marker conversion pipeline, supporting attachments, images, and audio/video input-to-text — designed for real work message flows.
Built-In Memory & Triggers
Memory, skills, and trigger mechanisms are deeply integrated into the runtime. Agents can autonomously evaluate trigger cycles and work proactively for you.
Rust High-Concurrency Foundation
Built on Rust for extreme stability and clear concurrency boundaries in long-running operations, effortlessly handling intensive Agent tasks.
A Different Path Altogether
Instead of a bigger gateway — a stronger collaboration structure
| Comparison Dimension | AG2O (A2) | Traditional AI Assistant |
|---|---|---|
| Core Narrative | Three-layer Agent collaboration system | Single assistant / chat window |
| Operation Mode | Persistent Daemon, always online | Temporary startup, on-demand use |
| Task Organization | Layered routing → specialized division | Dump all tasks into one session |
| Memory Accumulation | Independent workspace + persistent metadata | Simple session list search |
Let AI Actually Take Over Your Repetitive Work
A2 is a Personal Agent system that runs around you persistently. Let the top Agent understand you, let specialized Agents reside in their domains long-term, and let temporary execution units work in parallel.