Multi-agent AI platform

The problem

One general assistant is not enough for real work. I wanted specialists — each with its own memory, tools and permissions — that I can reach from my phone and that hand long jobs to background workers.

What I built

• An agent fleet on Telegram. Each agent has its own workspace, memory, skills and scoped credentials.
• Background workers. Agents delegate long-running tasks and report back when they finish.
• MCP servers and custom skills that give agents real tools, from media generation to publishing.
• Content automation: video pipelines with ffmpeg and ElevenLabs, publishing through the Facebook Graph API.
• Upstream contributions to claude-gateway, the multi-agent gateway the fleet runs on: computer use, MCP tools and orchestration.

Architecture

Telegram
   │
   ▼
claude-gateway        routing · memory · sessions
   │
   ▼
agent (Claude Code)
   ├── skills
   ├── MCP servers ──▶ media · publishing · browser
   └── background workers
          │
          └──▶ result back to the chat

Outcome

• 8+ specialized agents running day to day on a self-hosted gateway.
• Video and content automation with ffmpeg, the Facebook Graph API and ElevenLabs.
• Contributor to claude-gateway — 53 commits on computer use, MCP and orchestration.