<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Mingles Router — engineering blog</title><description>Building and running AI agents cheaply: OpenAI-compatible routing, agent runtimes, MCP tooling, real cost numbers.</description><link>https://router.mingles.ai/</link><language>en-us</language><item><title>We vibe-coded a farm of Telegram agents in 2–3 days — here are the guts</title><link>https://router.mingles.ai/blog/vibecoding-the-farm/</link><guid isPermaLink="true">https://router.mingles.ai/blog/vibecoding-the-farm/</guid><description>A dev diary of a Telegram agent farm: cheap inference as layer zero, Hermes as the runtime, MCP for tool security — including the rakes we hit.</description><pubDate>Thu, 25 Jun 2026 00:00:00 GMT</pubDate><category>telegram ai agents</category><category>vibe coding</category><category>cheap llm inference</category><category>mcp tools</category><category>agent runtime</category></item><item><title>One router, three agents: how our support, admin and marketing bots share one cheap inference endpoint</title><link>https://router.mingles.ai/blog/one-router-three-agents/</link><guid isPermaLink="true">https://router.mingles.ai/blog/one-router-three-agents/</guid><description>Support, admin and marketing bots on one shared brain: an OpenAI-compatible router, a Hermes runtime and an MCP tool layer — the full architecture.</description><pubDate>Thu, 18 Jun 2026 00:00:00 GMT</pubDate><category>ai agent architecture</category><category>openai compatible api</category><category>mcp server</category><category>hermes agent</category><category>telegram ai bot</category></item></channel></rss>