Run a 24/7 local agent that curates news digests
The author runs Hermes Agent on a Proxmox VM, a local Qwen 3.6 MoE curating news daily over MCP-connected RSS, delivered via Telegram and Signal. It runs at zero API cost, reportedly needing no supervision, though no accuracy or time-saved metrics are given. Task: The author deployed Hermes Agent on a Proxmox VM to orchestrate a local Qwen 3.6 MoE model running on an RTX 5060 Ti. The agent connects to RSS feeds and APIs via MCP, executes daily cron jobs to curate news, and updates a persistent memory file. Telegram and Signal serve as messaging gateways for delivery and private queries, while the human initially configured the cron schedule and data sources.; tools: Hermes Agent. Human role: The original post does not say how a person stayed involved.; limitations: Prerequisites include a local GPU capable of running the Qwen 3.6 MoE model and a Proxmox VM for the agent. The setup relies on self-hosted infrastructure, eliminating API costs but requiring maintenance of the llama-server and messaging gateways. Security concerns are noted, with the author recommending kernel-level sandboxing (nono/Landlock) for the autonomous agent to mitigate risks associated with unattended terminal access..
Tools
Published on Agent Work Atlas:
Overview
The author runs Hermes Agent on a Proxmox VM, a local Qwen 3.6 MoE curating news daily over MCP-connected RSS, delivered via Telegram and Signal. It runs at zero API cost, reportedly needing no supervision, though no accuracy or time-saved metrics are given.
Workflow
The author deployed Hermes Agent on a Proxmox VM to orchestrate a local Qwen 3.6 MoE model running on an RTX 5060 Ti. The agent connects to RSS feeds and APIs via MCP, executes daily cron jobs to curate news, and updates a persistent memory file. Telegram and Signal serve as messaging gateways for delivery and private queries, while the human initially configured the cron schedule and data sources.
Reported result
The system delivers structured daily briefings with zero API cost, leveraging 125 t/s inference speed to complete multi-step reasoning loops quickly. Over time, the agent's persistent memory accumulates patterns, transforming raw data into contextual insights like trend detection. The author reports that the agent now operates autonomously without active supervision, though specific quantitative metrics on accuracy or time saved are not provided.
Human role
The original post does not say how a person stayed involved.
Limitations
Prerequisites include a local GPU capable of running the Qwen 3.6 MoE model and a Proxmox VM for the agent. The setup relies on self-hosted infrastructure, eliminating API costs but requiring maintenance of the llama-server and messaging gateways. Security concerns are noted, with the author recommending kernel-level sandboxing (nono/Landlock) for the autonomous agent to mitigate risks associated with unattended terminal access.
Try it
I want to set up a local AI agent for daily news curation using Hermes Agent and a local LLM. First, verify I have a GPU capable of running Qwen 3.6 MoE and a Proxmox VM. Install Hermes Agent and configure it to connect to my local llama-server. Set up MCP connections for [replaceable RSS feeds or APIs]. Create a cron job for 8:00 AM that instructs the agent to fetch data, compare it with persistent memory, and generate a briefing. Configure Telegram as the delivery channel. Before enabling the daily cron, run a single manual test to verify the output format and data accuracy. Review the generated briefing and confirm it meets my needs. Only after successful manual verification should I enable the automated daily schedule. Ensure the agent runs in a sandboxed environment for security.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.