June 1, 2026
I now have two independent AI agents running on the same Dell Precision — each with its own Telegram bot, its own model, its own memory. Here's how.
Solomon (default profile) runs on DeepSeek V4 Pro. He's my primary assistant — cognitive auditor, execution-gated, friendly. He handles research, system administration, project work, and conversation.
DeepSeek R1 (r1 profile) runs on deepseek-reasoner. He thinks longer and deeper — designed for complex analysis where speed isn't the priority. Separate Telegram bot, separate gateway process.
Telegram bot tokens map 1:1 to Hermes gateway instances. To have two different bots with two different models, you need two gateway processes. I created a second systemd service with its own HERMES_HOME pointing to the r1 profile directory:
[Service] Environment="HERMES_HOME=/home/gershomj/.hermes/profiles/r1" ExecStart=...hermes gateway run --replace
Both agents read from ~/.hermes/shared/ — a directory of markdown files that serve as a single source of truth. User profile, system architecture, active projects. If one agent updates a file, the other sees it.
Every message from both bots feeds into a conversation vault — a SQLite database that tracks vocabulary, decisions, failures, and milestones. A cron job syncs every hour. The system gets smarter the more you use it.