An Agent That Writes Your Morning Brief While You Sleep: Automating Market Tracking

The video opens with a contrast: the narrator asleep at 07:58, a one-page message landing on his phone at 08:00. It gathers overnight Asian markets, the direction of US futures at the open, news on three portfolio companies and the day's earnings calendar on a single screen. Neither the narrator nor an assistant wrote it; it was written by an autonomous agent that did what it had been told the night before and finished the job.

The promise is clear: the same message should land on the viewer's phone too, set up from scratch, with one click and without hours in a terminal. But the narrator says upfront that setup is the easy part. The real issue is how the agent is told what to do. A badly briefed agent quickly turns into a bot firing dozens of notifications a day until the user mutes it, which is why the final section is devoted to instruction design.

The framing rests on two familiar references. First, the Buffett anecdote: asked about the secret of his success, he said he spends most of his day reading, so investing is largely a discipline of gathering information, not raw intelligence. Second, Housel's noise thesis: the investor's enemy is not lack of information but the inability to separate signal from noise. Anyone who burns twenty morning minutes on social-media arguments and comment threads and walks away with two or three real facts is already living the problem this video targets.

A key distinction follows: Hermes is not a chat agent but an autonomous agent. A chatbot works when asked; an agent works while the user is busy elsewhere. Three capabilities make it an agent: it can carry multi-step work through on its own, it reads live web pages instead of reciting memorized knowledge, and it can run scheduled tasks. The morning briefing combines all three: collect data overnight, summarize by morning, deliver on schedule.

The video explains in detail why the system goes on a server, not a laptop. When the laptop sleeps or the home connection drops, the agent sleeps too and scheduled tasks never run. The fix is an independent, always-on machine with its own connection: a virtual private server. The sponsor, Hostinger, is recommended on three grounds: a one-click app catalog with Docker templates, round-the-clock support for first-timers, and a 30-day refund. The KVM2 plan is picked for this workload; two virtual cores, 8 GB of memory, 100 GB of disk and 8 TB of traffic are judged enough to carry the agent itself plus the data layer and page-scraping jobs at once.

Setup is designed for a non-technical viewer. A server location is chosen, Hermes is found in the app catalog and the Docker-templated agent product is installed; the panel opens without a single terminal command typed. The narrator adds two practical warnings: write the panel password down somewhere, because the system runs itself and the panel may go unopened for weeks; and extra API keys can optionally strengthen the agent. A short experience survey follows, then login to the interface.

The heart of the video is the morning-briefing workflow. A job firing daily at 08:00 is told to scan Asian markets, the European open, US futures, the dollar and gold, plus general macro headlines. But the output format is ruled as strictly as the coverage: at most ten items, one line each, no commentary and no forecasting. Without that rule the agent sends a three-page essay nobody reads in the morning. Model choice is a separate step: an existing chat subscription can be connected, and search keys such as Brave can be added for deeper web coverage.

Telegram, not email, is chosen for delivery, on the grounds that a morning report drowns between invoices and marketing mail. Setup starts with opening a new bot via BotFather: a display name, a technical username ending in bot, the issued token and the chat identifier entered into the panel. For safety, only the owner's and the bot's identifiers go on the allowed-users list; then the bot restarts and is assigned as the home channel. A test brief sent and seen in Telegram counts as verification, and the schedule and channel settings are double-checked in the jobs list.

The video states plainly who the system is not for: anyone checking a single index once a day will find it over-engineered; the phone app's notification is enough and a monthly fee would be waste. It pays off for people tracking several markets, holding more than a handful of positions and spending their mornings hunting news. A boundary is drawn too: the system gives no buy or sell calls, it only collects and organizes; it answers the way it was configured. Never blindly acting on an AI agent's investment advice is one of the video's direct warnings.

The close is a four-item security checklist: a strong panel password with two-factor authentication, no publicly exposed panel, no fiddling with unneeded ports and settings, restricted chat permissions on the Telegram side, and no personal data shared with bots. The narrator finishes by noting the setup was shown in its simplest possible form, with extra keys and model connections left as options for broader search or tidier prose.