--- title: "Build a Daily AI Research Digest with Hermes on Telegram" description: "Build a scheduled Hermes assistant that summarizes research sources and delivers a daily digest to Telegram — AI credits included, no API keys needed." canonical: https://agentroost.app/en/blog/ai-research-digest-summaries-hermes-telegram date: 2026-05-14T12:00:00Z --- [Canonical URL](https://agentroost.app/en/blog/ai-research-digest-summaries-hermes-telegram) Most researchers and marketers face the same quiet tax: every morning, thirty browser tabs, five newsletters, a couple of RSS feeds, and a Slack full of links. By the time you have read through it all, an hour has gone. This guide walks you through building a Hermes assistant on AgentRoost that handles the reading for you. You define the sources and schedule; Hermes fetches, summarizes, remembers what it covered yesterday, and pushes a tidy digest to your Telegram — powered by included AI credits you never have to top up. --- ## What You Are Actually Building A **persistent Hermes assistant** that: 1. Runs on a fixed schedule (daily, twice-daily, or weekday-only — your call). 2. Pulls content from sources you specify: RSS feeds, newsletter URLs, documentation pages, GitHub release notes, anything with a public URL. 3. Uses the included LLM to summarize each source into 3–5 key points. 4. Bundles the summaries into a single digest message and sends it to your personal Telegram bot — the one AgentRoost auto-provisions when you create the workspace. 5. Keeps a memory log so the next run skips items already covered, avoiding duplicate briefings. You do not write Python. You do not paste an API key anywhere. You configure the assistant in natural language and a short YAML file — and it runs 24/7 on AgentRoost infrastructure. --- ## Prerequisites - An AgentRoost account (sign up at [agentroost.app](/en/agents/hermes) with email, Google, Microsoft, or Discord). - Telegram installed on your phone or desktop. - The URLs of 3–10 sources you want to track. Public URLs work best; paywalled content will not parse. --- ## Step 1 — Launch Your Hermes Workspace 1. Log in and go to **Workspaces → New workspace**. 2. Choose the **Hermes** framework. 3. Give it a name — something like `research-digest`. 4. Click **Create**. Within about two minutes the workspace is live. At the end of provisioning, the manager bot sends you a message in Telegram with a `/start` link for your new agent bot. Click it. That is the channel where your digests will arrive. --- ## Step 2 — Configure Your Sources In the Hermes workspace editor, open (or create) `config/digest.yaml`: ```yaml digest: name: "Morning Research Brief" schedule: "0 7 * * 1-5" # 07:00 UTC, weekdays only memory_days: 7 # remember covered items for 7 days sources: - name: "Hacker News Top" url: "https://news.ycombinator.com/rss" type: rss max_items: 5 - name: "OpenAI Blog" url: "https://openai.com/news/rss.xml" type: rss max_items: 2 - name: "My custom page" url: "https://example.com/updates" type: html css_selector: "article.post" max_items: 3 output: format: telegram style: bullets # "bullets" or "paragraphs" max_length: 3000 # chars per digest message ``` Key fields: - `schedule`: standard cron syntax. `0 7 * * 1-5` = 7 AM UTC Monday–Friday. - `memory_days`: Hermes keeps a rolling log of item URLs it has already sent. Items seen within this window are skipped. - `type: rss` parses Atom/RSS feeds. `type: html` fetches a page and applies the CSS selector. - `max_items`: how many entries to pull per source per run. --- ## Step 3 — Write the Summarization Prompt Create `prompts/summarize.md`: ```markdown You are a concise research assistant. Given the following article, extract the 3 most important points a practitioner needs to know. Format as a tight bulleted list. No filler sentences. If the item is not relevant to AI, automation, or developer tooling, say "Skip — outside scope" and nothing else. Article title: {{title}} Article content: {{content}} ``` Hermes passes each fetched item through this prompt using the **included AI credits**. You pick the model from workspace settings — there are 350+ available models and you can switch anytime without touching billing. --- ## Step 4 — Wire the Assembly Prompt Create `prompts/assemble_digest.md`: ```markdown You are assembling a morning research digest for a busy professional. Compile the summaries below into a single Telegram message: 1. Start with "Today's brief — {{date}}". 2. Group summaries under their source name as a bold header. 3. Keep each source block to 5 lines maximum. 4. End with: "{{n}} items · {{skipped}} skipped as duplicates". Summaries: {{summaries}} ``` The `{{}}` variables are filled in by Hermes's template engine before the prompt hits the LLM. --- ## Step 5 — Test and Go Live In the workspace terminal (accessible from the AgentRoost dashboard), run a dry-run first: ```bash hermes task run digest --dry-run ``` This fetches sources and renders the digest without sending anything to Telegram. Check each source's output and tweak `max_items`, CSS selectors, or prompt tone until you are happy. Then send a real message: ```bash hermes task run digest ``` After the first real run, Hermes writes a `memory/digest_seen.jsonl` file — one entry per item URL with a timestamp. On the next scheduled run, duplicate items are filtered automatically and the `{{skipped}}` count in your digest tells you what was removed. --- ## Tips for the First Week - **Multiple digests, one workspace**: copy `digest.yaml` to `digest-evening.yaml` with a different schedule and source list (competitor blogs, product announcements). Each config runs on its own cron independently. - **Vary the model by task**: use a lighter model for per-item summarization (faster, uses fewer credits) and a more capable model for the final assembly step. Set this per prompt file in the Hermes config. - **Cron timezone**: AgentRoost runs schedules in UTC. If you want 8 AM Istanbul time (UTC+3), set `0 5 * * *`. - **Paywalled or JS-heavy pages**: `type: html` does a plain HTTP fetch. If the source requires login or client-side rendering, the content will be empty. Prefer RSS where available. --- ## How to Do This on AgentRoost Hermes on AgentRoost costs from **$19.99/mo all-in**. That price includes the server, the always-on process, and the AI/LLM credits powering every summarization run — there is no separate API key to buy or manage. Every competitor (Elestio, Sliplane, self-hosted) makes you bring your own OpenAI or Anthropic key and pay that bill on top of hosting. On AgentRoost the credits are already in the subscription. The Telegram bot is auto-provisioned the moment the workspace starts. You do not visit BotFather, generate a token, or configure a webhook URL. The connection is already set up. **The real journey:** 1. [Create a Hermes workspace](/en/agents/hermes) — about 2 minutes. 2. Add `config/digest.yaml`, `prompts/summarize.md`, and `prompts/assemble_digest.md` via the file editor. 3. Click the `/start` link in Telegram when the bot message arrives. 4. Run `hermes task run digest` in the terminal to verify the output. 5. The scheduler takes over. Your digest arrives every morning. [Get started — Hermes plans from $19.99/mo](/en/pricing) · [Compare all frameworks](/en/agents) There is a 14-day money-back guarantee and no annual commitment required.