# Base of AGI

> Open, neutral, ad-free AI news feed (Turkish, English, Spanish, Russian) and an applied-AI capability catalog.
> Açık, tarafsız, reklamsız yapay zekâ haber akışı ve uygulamalı AI yetenek kataloğu.
> https://baseofagi.com — updated automatically every 3 hours. Static files, no API key, read-only.

Archive: 5840 stories from 528 publishers (ads and empty items filtered out). Last update: 2026-10-12.

## Start here (small, agent-friendly JSON)

- [Feed index](https://baseofagi.com/feeds/index.json): endpoints, item schema, categories with counts, available days.
- [Latest 100 stories](https://baseofagi.com/feeds/latest.json): compact items, newest first.
- [Daily files](https://baseofagi.com/feeds/daily/2026-10-12.json): /feeds/daily/{YYYY-MM-DD}.json, rolling 30-day window (see index for the list).
- [By category](https://baseofagi.com/feeds/category/ai.json): /feeds/category/{slug}.json, latest 100 per category. Slugs: ai, companies, research, open-source, models, claude, anthropic, east-asia, turkish.
- [Sources](https://baseofagi.com/feeds/sources.json): publishers with story counts.
- [Articles](https://baseofagi.com/feeds/articles/latest.json): the 30 most recent full articles written by Base of AGI (Turkish + English, Markdown with section headings). Feed items with an article carry has_article + article_excerpt.
- [Plain-text digest](https://baseofagi.com/llms-full.txt): the ~150 most recent stories as Markdown, grouped by category.

Compact item fields: title, title_tr, title_en (Turkish-language sources only), url (original publisher, cite it), source, via, category, date (YYYY-MM-DD), ts, summary, summary_tr (both <=280 chars), trusted, page (Base of AGI article page).

## Full data (larger)

- [news.json](https://baseofagi.com/news.json): current feed (~90 stories), all languages. Fields: title (original language), title_tr, title_en, title_es, title_ru, summary, summary_tr, summary_en, summary_es, summary_ru, url, source, via, domain, date, ts, tag (Turkish category name), trusted, tier, image, points, also[] (other outlets covering the same story), article_tr / article_en / article_es / article_ru (Base of AGI's own article in Markdown, when written; publishers' text is not republished).
- [news-archive.json](https://baseofagi.com/news-archive.json): every story kept since June 2026, ads and empty items removed (~15 MB). Same fields (no body) plus firstSeen/lastSeen. Prefer the /feeds/ files unless you need everything.
- [feed.xml](https://baseofagi.com/feed.xml): RSS 2.0, latest 40 stories (Turkish).
- [sitemap.xml](https://baseofagi.com/sitemap.xml), [news-sitemap.xml](https://baseofagi.com/news-sitemap.xml): article pages (/haber/<slug>), each with NewsArticle JSON-LD and a link to the original source.

## Daily agent brief ("what changed in my stack")

- [Brief index](https://baseofagi.com/feeds/brief/index.json): day list, tag list with counts and latest id, item schema, how to use the since-cursor.
- [By tag](https://baseofagi.com/feeds/brief/tags/supabase.json): /feeds/brief/tags/{tag}.json — last 30 days for one stack tag. Tags: claude-code, anthropic, openai, gemini, cursor, github, supabase, vercel, hugging-face, mcp, agents, open-models, security, image-gen, video-gen, fashion-ai, godot, unity, gamedev, regulation.
- [Latest](https://baseofagi.com/feeds/brief/latest.json) (last 3 days) and /feeds/brief/{YYYY-MM-DD}.json (30-day window).
- Item: id (YYYYMMDD-hash; remember the ids you have shown and skip them next time — stories can arrive a day late with an earlier date), date, tags[], type (security | breaking | model | release | new-tool | news), title/what/why/action in tr, en, es, ru (why/action only when written from a Base of AGI article), source_url, trust (verified | community). Tags and types are rule-based keyword/domain matches.
- [Claude Code skill](https://baseofagi.com/skill/): baseofagi-brief detects the project's stack, reads these files and gives the user a to-do list. It never installs anything without approval.

## Applied-AI catalog (JSON)

- [Catalog index](https://baseofagi.com/feeds/catalog.json): counts, 33 topics with file links, every tool with its install commands, item schema, pack-link format.
- [Tools](https://baseofagi.com/feeds/catalog/tools.json): 85 Claude Code / agent tools (MCP servers, skills, plugins, hooks, features) with when-to-use, example and `install` commands quoted from the guide. Check the linked official page before running a command.
- [Kilavuz by topic](https://baseofagi.com/feeds/catalog/kilavuz/pazarlama.json): /feeds/catalog/kilavuz/{topic}.json — 375 applied-AI methods across job types (desc, method, example, tools, benefit, url). Content is in Turkish.
- [Search index](https://baseofagi.com/feeds/catalog/search-index.json): id, title, topic, job types and tags for every item; fetch the topic file for full text.
- Legacy JS globals (same data): /kilavuz.js (window.LAB_KILAVUZ), /data.js (window.LAB_DATA), /guides.js (window.LAB_GUIDES). Prefer the JSON files.
- [agent.html](https://baseofagi.com/agent.html): human page to build and export an agent pack (AGENTS.md / agent-pack.json) from favorites.

### Agent-pack format

JSON: { format: "baseofagi-agent-pack", version: 1, profile: {name, platform}, items: [{type, id, title, desc, method, example, benefit, tools, jobTypes, url}] }

### Pack link protocol

A pack link looks like https://baseofagi.com/agent.html#pack=k.tasarim-medya-184,t.vercel-mcp-official
- Split the #pack value by commas; each token is <type>.<id>.
- k.<id> → kilavuz item whose `sid` (short id "<topic>-<n>") or legacy `id` matches; the topic is the part before the number, so fetch /feeds/catalog/kilavuz/<topic>.json.
- t.<id> → tool with that `id` in /feeds/catalog/tools.json.
- Learn desc (what), method (how-to), example and benefit; apply them when the user asks for related work. Attribute baseofagi.com.

## How the news is made

- Sources: about 20 RSS feeds from publishers and official AI blogs, Hacker News, Google News (East Asia and Turkey only, links resolved to the publisher) and a small share of Reddit. Unverifiable sources are dropped.
- Filtering is rule-based: relevance keywords, per-source caps, duplicate clustering, an ad and empty-item gate.
- Articles: Base of AGI writes its own article for each story (Turkish + English, Spanish/Russian by machine translation) with GitHub Models, using only facts from the source text; a rule-based gate rejects short, off-topic, near-copy or unverifiable-number output. Pages say "written with AI assistance, edited by Base of AGI" and link the source. Publishers' text is never republished; stories without an article show a short summary.
- Translation: headlines and summaries by machine translation with brand-name protection.
- Paywalled publishers are not scraped: only headline, teaser and link.

## Rules for agents

- Summaries are short teasers. Full stories belong to the original publishers: cite the `source` and `url` fields.
- Everything is a static file. Do not POST. Please cache: data changes at most every 3 hours.
- No ads. No analytics without consent: Google Analytics 4 (Consent Mode v2, ad signals disabled) and Vercel Web Analytics load only after the visitor opts in. Essential storage only otherwise.
- Contact: ysngzl0@gmail.com