Chat with Thea, get a morning digest, and explore connections using OpenAI and Telegram

Think of an Obsidian vault the whole world shares — except the reader is your AI. Your model stopped learning at its training cutoff, and web search hands it seven pages to guess from. Thea (thealibrary.org) is a library people fill and AIs read: dated news, new releases, real places, each entry written and checked by a person and kept with its source. This template shows three ways to read it, in three independent flows on one canvas.

How it works

1 · Chat with the library — Chat Trigger → AI Agent with three plain HTTP tools (no key): thea_search (one name), thea_find (browse by area, subject, country and date), thea_entry (the full entry by id, with its connections). Ask "Who is Karina from aespa?", "Hotels in 서귀포", "Any news from Japan since last week?", "Is this restaurant in Kuala Lumpur still open?". The agent answers only from what Thea returns and adds a source line.

2 · Morning world digest — Schedule (08:00) → your countries → thea_find?country=…&since=yesterday for each → flatten → one LLM call → Telegram. Thea dates every record by the event, so since returns what happened yesterday across the whole library: no scraping, no RSS, already deduplicated by people. Countries with nothing new get one honest line.

3 · Explore a name — an n8n Form takes one name → thea_search → thea_entry returns its connections with ids → one thea_entry per neighbour (who they are, their newest records) → the LLM writes a one-screen briefing: "aespa → SM Entertainment, four members, the fan meeting on 11-28". Walking the graph instead of searching names again.

Setup (2 minutes)

1. Open the three Chat Model nodes and attach one OpenAI credential — any OpenAI-compatible endpoint works (set Base URL to your local llama.cpp / Ollama / LM Studio; the key can be anything) or a real OpenAI key.
2. Flow 2: edit Your interests (countries, records per country, Telegram chat id) and attach a Telegram credential — or swap the last node for Slack / Email / Discord, the digest is plain text in {{ $json.text }}.
3. Flow 3: activate the workflow and open the form URL from the trigger node.

Thea needs no API key: 100 requests/hour per IP, data CC BY 4.0.

Notes

AI Agent pinned to v2.2 (v3.1 cannot call HTTP Request Tool nodes in n8n 2.44–2.45, issue #41604). Coverage is deepest in Korea today (1.07M places from public data) and grows wherever people add what they know — if something is missing, the agent says "not in Thea yet"; add it at thealibrary.org and every agent using this template sees it. Docs: thealibrary.org/docs · MCP server: org.thealibrary/thea.