AIHOT on Railway
Self-updating AI news site with scored items, grouped events and a daily report, on Railway.
AIHOT is an open-source news site that runs itself. It collects items from the sources you list, has a language model screen and score each one twice, writes a headline and summary, groups reports of the same story into one event, and ranks events by how many sources discuss them. It then publishes a daily report, with weekly and monthly roundups.
People use it to run a news or briefing site for a field of their own: change the sources, scoring prompts and thresholds in a fork and the same pipeline covers another industry. It also exposes RSS, a public JSON API, an MCP server and an llms.txt file, so agents can read it as well as people.
The interface and everything the site writes are in Simplified Chinese, which is upstream's design. AIHOT is MIT-licensed; the name and logo are not covered by that licence, and upstream asks sites to use their own.
What AIHOT can do
- Collects items from RSS-style sources, with 18 English AI news feeds as demo sources
- Two-pass model screening and scoring, then headline and summary writing
- Groups reports of the same story into events and ranks them by heat
- Daily, weekly and monthly reports and a trending list
- RSS, public JSON API, MCP server and llms.txt
- Admin page with sources, run history and model usage per step
What people use it for
- A self-updating AI news site with a daily report
- A news site for your own field, built from a fork with your sources and prompts
- A curated feed for AI agents and readers
- A Chinese-language AI digest for readers and teams who read Chinese
Videos
Running AIHOT on Railway
The Railway template deploys a Caddy gateway, the web site, a backend that runs the API and background worker, and PostgreSQL 17, plus a volume for images and uploads and a Railway Bucket that receives daily backups. It builds a pinned upstream commit, checked against a SHA-256 checksum, because upstream has no releases. You supply an API key for an OpenAI-compatible model service (DeepSeek is the default) and the admin password is generated. Cost comes from the Railway services running all day and from model calls billed by your provider.
AIHOT templates on Railway
Tracklayer templates are listed first. Other AIHOT templates follow, ordered by total deployments on Railway. How this works.
Deployment numbers as of 2026-11-20.
Questions about AIHOT
What do I need before deploying?
An API key for an OpenAI-compatible model service, pasted into LLM_API_KEY. DeepSeek is the default; OpenAI, OpenRouter, Qwen and others work after changing LLM_BASE_URL and LLM_MODEL. The provider bills model calls.
Can I get the site in English?
Not through variables. The interface and generated text are Simplified Chinese by upstream's design. Language, site name, report times and prompts live in the code, so you would fork the repository, edit it and point the template at your fork.
How do I sign in to the admin?
Open /admin on your domain. The password is ADMIN_PASSWORD in the Backend service variables.
When does the first daily report appear?
At the first 08:00 China Standard Time (00:00 UTC) after you deploy. Until then the admin Runs page showing nothing judged for reports is expected.
What drives the cost?
Model calls grow with the number of new items per day; the admin shows calls and tokens per step. Setting MODEL_CALLS_ENABLED to false stops model spending. Backups go to a bucket daily, so delete old ones from time to time.
Similar apps
Tracklayer maintains the templates marked as such; AIHOT is a trademark of its owners and this site is not affiliated with them.