Cláudio: An Autonomous Agent Operating System Based on Telegram

Title: Cláudio: An Autonomous Agent Operating System Based on Telegram

Abstract: Cláudio is an autonomous agent operating system developed in Go, which provides a natural interaction interface via Telegram and supports multi-agent collaboration and local large language model (LLM) infrastructure.

Original Author/Maintainer: JConradoN Source Platform: GitHub Release Date: June 2, 2026 Original Link: https://github.com/JConradoN/claudio

Core Features: Lightweight and cross-platform, operating system design philosophy (persistent services, memory and task scheduling), multi-agent workflow support, local LLM infrastructure.

Project Background and Positioning

Cláudio is an autonomous agent operating system developed in Go, designed to provide users with a lightweight, cross-platform intelligent assistant infrastructure. It is a research branch of the Aurelia project, retaining the advantages of the original architecture while expanding support for multi-agent workflows and local large language model (LLM) infrastructure.

Unlike traditional chatbots, Cláudio adopts an operating system design philosophy, treating agents as persistent services that can run long-term and have memory and task scheduling capabilities. Users interact via Telegram and can use it without additional software installation.

Core Architecture Design

Cláudio uses a layered design:

• Bottom layer: PI SDK inference engine (responsible for core functions such as model calling, tool execution, and session management)
• Middle layer: Aurelia product layer (covers user-oriented functions like identity management, personalized settings, Telegram user experience, workflow orchestration, and persistent memory)
• Top layer: Interaction interfaces (Telegram, command line, scheduled tasks, and future expansion interfaces)

Layered advantages: Clear responsibilities—PI SDK focuses on core capabilities, Aurelia focuses on product experience and continuity assurance; when PI SDK has existing capabilities, Aurelia adapts and orchestrates; product-specific memory/strategies/workflow states are managed by Aurelia as the only trusted source.

Technology Stack and Dependencies

• Core daemon process: Developed in Go (using lightweight goroutines for efficient concurrency), requires Go 1.25+
• Dependencies: Node.js 18+, npm 8+ (supports PI SDK operation)
• Data persistence: SQLite database (reliable and easy to deploy)
• Telegram integration: Official Bot API (users need to obtain a token from BotFather to enable the message interface)
• Model support: Multi-provider compatibility (OpenRouter, opencode-go, Anthropic, Kimi, Z.ai, Alibaba, etc.)

The technology selection balances performance and maintainability.

Execution Model and Capability Boundaries

Key components of the execution model:

• 24/7 running Go daemon (basic service)
• TypeScript bridge layer (encapsulates PI SDK to enable interaction with the inference engine)
• PI coding agent (the system's "brain", responsible for task planning, tool calling, and session management)

Core capabilities: Natural dialogue interaction, task delegation decision-making, automated execution scheduling, persistent memory management, multi-project support. When receiving user messages, it intelligently decides to respond directly, delegate to a professional agent, or arrange a scheduled task.

Memory system features: Supports scope memory and automatic information extraction, maintains context continuity across sessions; supports Markdown configuration files and scheduled task scheduling for easy customization of agent behavior.

Use Cases and Value Proposition

Applicable scenarios:

• Individual users: All-day intelligent assistant (daily queries, task reminders, information organization)
• Developers: Extensible agent framework (supports custom tool and skill development)
• Research teams: Multi-agent collaboration capabilities (infrastructure for complex task decomposition and parallel processing)

Value proposition: Compared to direct use of LLM APIs, it adds essential production environment capabilities such as persistence, memory, scheduling, and multi-project support; reduces usage barriers via Telegram—users do not need to care about underlying model calls or session management, and can complete tasks through natural dialogue.

Future Development Directions

Long-term differentiation goal: Wiki Memory Gateway—a local-first, Markdown-auditable, scope-isolated memory layer. This layer can be directly called by Aurelia, PI, PI Code/opencode, and future MCP clients, ensuring data isolation between different users and projects.

The design idea reflects attention to data sovereignty and privacy—users can fully control data and memory locally without relying on cloud services. As LLM technology evolves, local-first and auditable architectures may become an important development direction for agent systems.