vibex — Architecture

vibex bridges Telegram Forum topics and a web UI to AI coding assistants (Claude, Codex, Gemini). Single-agent chats, multi-agent group collaboration, scheduled tasks, and MCP orchestration — all on the same platform. Click any diagram below to explore.

☰

System Overview

Full architecture map — all layers from user interfaces through application logic, MCP servers, AI backends, down to persistence.

High-Level

✉

Web Chat Flow

Single-agent conversation: Browser → FastAPI → AgentExecutor → CLI subprocess → JSONL → SSE streaming.

Sequence Diagram

☆

Group Chat

Multi-agent collaboration — Claude, Codex, Gemini working together on the same CWD via MCP tools, orchestrator, watchdog.

Most Detailed

✈

Telegram Bot

Topic ↔ tmux window ↔ session mapping. Message delivery, JSONL polling, interactive UI, directory browser.

Bidirectional Flow

⚙

MCP Server

vibex-subagents on :7891 — run_agent, agent_result, sessions. Non-blocking background task execution with parallel support.

3-Tool API

⏰

Scheduled Tasks

Cron-based scheduler — polls every 30s, creates conversations, runs agents, broadcasts SSE, prunes history.

Automation

✎

Backend Abstraction

Backend ABC, registry, command building, JSONL parsing, token extraction. Claude vs Codex vs Gemini comparison.

3 Implementations

◫

Database Schema

14 SQLite tables — conversations, messages, tool_calls, group chats, agent definitions, MCP servers, scheduled tasks.

ER Diagram

⚡

Domain Events & SSE

Event pipeline: Backend JSONL → ParsedEntry → DomainEvent → SSE. Full event catalog with data fields.

Event-Driven