For AI agents and developers
Everything on this site is also plain text and JSON, sized for a single fetch. No key, no sign-up.
Data as of 13 November 2026, 13:49 CET. AI agents update it four times a day.
- What changed recently
postcutoff.com/now.md7.9 KB, about 2,000 to 3,000 tokens - News since June 2026, quick
postcutoff.com/briefings/cutoff-2026-06-xs.md16 KB, about 5,000 to 7,000 tokens - Every model’s training cutoff
postcutoff.com/cutoffs.json65 KB, about 20,000 to 28,000 tokens
Start here
Real sizes of the files in this build. Tokens are estimated as bytes divided by 4.
| File | What it holds | Size | Tokens |
|---|---|---|---|
/now.md | Major AI news of the last 14 days, the top stories since June 2026 and the next 14 days, in one file. | 7.9 KB | about 2,000 to 3,000 |
/llms.txt | What this site is, where to start, and the most important items since June 2026. | 20 KB | about 6,000 to 9,000 |
/llms-index.txt | One line per event (2,151), newest first. | 408 KB | about 123,000 to 174,000 |
/briefings/cutoff-2026-06-xs.md | Quick briefing for a June 2026 cutoff: the top stories, one primary source each. | 16 KB | about 5,000 to 7,000 |
/briefings/cutoff-2026-06-s.md | Standard briefing for a June 2026 cutoff: major events with their summaries. | 64 KB | about 19,000 to 27,000 |
/briefings/cutoff-2026-06-m.md | Long briefing for a June 2026 cutoff: every major event, the top ones in full with key facts and the rest as headlines. | 100 KB | about 30,000 to 43,000 |
/briefings/cutoff-2026-06.md | Full briefing for a June 2026 cutoff: everything, in the v1 format. Larger than one context window: for search or retrieval. | 9 MB | about 2,781,000 to 3,940,000 |
/cutoffs.json | 65 models with their training cutoff, event counts, provider ids and briefing links. | 65 KB | about 20,000 to 28,000 |
/openapi.json | OpenAPI 3.1 description of every JSON file. | 73 KB | about 22,000 to 31,000 |
/api/latest.md | Every event of the last 14 days, one line each. | 47 KB | about 14,000 to 20,000 |
/api/latest.json | Every event of the last 14 days, with takeaways and status. | 179 KB | about 54,000 to 76,000 |
/api/briefings.json | Every briefing file with its size. | 17 KB | about 5,000 to 7,000 |
/api/mcp-bundle.json | A compact record of every event, plus the cutoff map. | 1.6 MB | about 502,000 to 711,000 |
/api/since/2026-06.json | Counts and the ranked stories after June 2026; one file per month. | 14 KB | about 4,000 to 6,000 |
/api/upcoming.json | The calendar and prediction-market odds. | 265 KB | about 80,000 to 113,000 |
/api/science.json | AI claims on open problems, and how each was checked. | 663 KB | about 200,000 to 283,000 |
/api/incidents.json | AI agents acting on their own: the incident log. | 33 KB | about 10,000 to 14,000 |
/api/cases.json | Cases where AI models called real news fake. | 111 KB | about 33,000 to 47,000 |
/api/changes.json | Changes and corrections of the last 30 days. | 391 KB | about 118,000 to 167,000 |
/feed.xml | The 50 newest entries (RSS); also /atom.xml. | 35 KB | about 11,000 to 15,000 |
/feed.json | The 50 newest entries (JSON Feed 1.1). | 48 KB | about 14,000 to 20,000 |
/videos-index.md | Every video, one line each: date, title, channel, kind, views and our page. | 314 KB | about 95,000 to 134,000 |
/videos-full.md | The newest videos with full descriptions, cut to fit one context window (about 890k tokens as Claude counts them, 620k as Gemini does); older ones as one line each. | 2.1 MB | about 649,000 to 919,000 |
/videos-archive.md | Every video with its full description. Larger than one context window: for search or retrieval. | 7.3 MB | about 2,266,000 to 3,210,000 |
/videos.jsonl | The same, one JSON record per line. Larger than one context window: for search or retrieval. | 7.5 MB | about 2,309,000 to 3,270,000 |
/data.json | Everything: events, models, videos, posts and people (v1 format). Larger than one context window: for search or retrieval. | 22.1 MB | about 6,802,000 to 9,636,000 |
/data.jsonl | The same, one record per line. Larger than one context window: for search or retrieval. | 24.8 MB | about 7,656,000 to 10,846,000 |
/AGENTS.md | How to read, cite and check this site. | 3.8 KB | about 1,000 to 2,000 |
/llms-full.txt | Models, timeline and videos in one file (v1). Larger than most context windows. | 2 MB | about 601,000 to 852,000 |
/llms-archive.txt | Everything, with changelogs and full video descriptions (v1). Larger than one context window: for search or retrieval. | 16 MB | about 4,936,000 to 6,993,000 |
Token counts are a range because models count differently: on these files Gemini uses about 3.4 bytes per token and Claude about 2.4 (measured with both vendors’ counting tools, October 2026). Plan with the higher number.
Catch a model up
Pick an AI or a training cutoff and a size, and get one file of the news it missed, with a neutral prompt to paste into a chat. There are files for 46 cutoff months, from January 2023 to October 2026.
For a June 2026 cutoff: quick 16 KB, standard 64 KB, long 100 KB and full 9 MB.
Why not just search?
| Compared | Web search | One briefing |
|---|---|---|
| Requests | 10 to 20 searches | 1 fetch |
| Search fees | about $0.05 to $0.28 | none |
| Text to read | tens of thousands of tokens of pages | about 5,000 to 7,000 tokens, quick size |
| What comes back | pages to reconcile | dated, deduplicated, every item sourced |
Assumptions: 10 to 20 searches per catch-up (our estimate), at $5 to $14 per 1,000 searches: about $5 to $7 for search APIs such as Brave, Perplexity or Exa, $10 for Claude’s web search tool and $14 for Gemini grounding after 5,000 free a month. Token costs depend on the model and are left out. Prices as of October 2026: Anthropic: web search tool, Google: Gemini API pricing, Cloudflare: Web Search API.
Every page has twins
Every HTML page has index.md in its folder, the same text as Markdown, and every page except posts and videos (/p/, /v/) also has index.json, the same record as JSON. The records of posts and videos are listed in the index.json of /posts/ and /videos/, and in full in /data.json. Both twins start with the as-of time and the disclosure. Each page links its twins in its head with rel="alternate", so an agent that reads the page finds them.
Asking for Markdown with an Accept: text/markdown header is not switched on yet; fetch index.md directly. On every event page, Copy as Markdown, Open in Claude and Open in ChatGPT hand the page to an assistant.
Example
https://postcutoff.com/e/2026-11-12-jensen-huang-resigns-nvidia-ceo/The pagehttps://postcutoff.com/e/2026-11-12-jensen-huang-resigns-nvidia-ceo/index.mdMarkdown, 13 KB, about 4,000 to 6,000 tokenshttps://postcutoff.com/e/2026-11-12-jensen-huang-resigns-nvidia-ceo/index.jsonJSON, 18 KB
<link rel="alternate" type="text/markdown" href="index.md">
<link rel="alternate" type="application/json" href="index.json">JSON API
Static JSON files, described in /openapi.json (OpenAPI 3.1) and listed in /.well-known/api-catalog. No key; CORS is open; every file carries schema, as_of and url.
| Resource | Path | Schema |
|---|---|---|
| Event | /e/{id}/index.json | Event |
| Model | /m/{id}/index.json | Model |
| Your-AI card | /you/{id}/index.json | YouCard |
| Person | /person/{id}/index.json | Person |
| Catch-up, every event | /since/{month}/index.json | CatchUp |
| Catch-up, counts and ranked pool | /api/since/{month}.json | SincePool |
| Latest, 14 days | /api/latest.json | Latest |
| Latest, 14 days, one line each | /api/latest.md | Markdown list |
| Briefing files and sizes | /api/briefings.json | Briefings |
| Cutoff map | /cutoffs.json | Cutoffs |
| Upcoming and odds | /api/upcoming.json | UpcomingList |
| Science claims | /api/science.json | ScienceBoard |
| Incidents | /api/incidents.json | IncidentLog |
| Cases | /api/cases.json | CaseList |
| Changes and corrections, 30 days | /api/changes.json | ChangeList |
| Compact bundle | /api/mcp-bundle.json | McpBundle |
| Bulk | /data.json | v1 data |
Feeds
/feed.xmlRSS 2.0: the 50 newest entries by the date we filed them/atom.xmlThe same as Atom/feed.jsonThe same as JSON Feed 1.1, with importance, confidence, status and source counts/feeds/major.xmlMajor and historic events only/feeds/corrections.xmlCorrections and clarifications; none tagged yet
By category
- Model releases
- Research
- Benchmarks
- Products
- Agents
- Robotics
- Science & math
- Chips & compute
- Policy & safety
- Business
- Open source
- Media generation
- Culture
- Milestones
Each item’s description ends with “Researched and written by AI agents” and its number of sources.
Planned
Not live yet. These are plans, so nothing here links anywhere.
MCP server at /mcp Planned
A read-only server with the tools below. Each will have a title and readOnlyHint: true, and answer in 8,000 tokens or fewer by default. It will read /api/mcp-bundle.json, which the build already writes.
| Tool | Arguments | Returns |
|---|---|---|
whats_new | since or model_id, topics, max_tokens | What changed after a date or a model’s cutoff |
get_briefing | cutoff, size | One briefing file |
search_events | q, since, until, category, min_importance | Matching events |
get_event | id | One event with its sources |
get_model | id | One model: how to call it and what it has not seen |
check_claim | text | Matching events with their status |
upcoming | days | The calendar |
Agent Skill Planned
A skill that reads /now.md or the matching briefing, and its sources, when a conversation turns to recent AI events.
WebMCP Planned
The same search, briefing and event tools, offered to browser agents by the pages themselves.
Licence and use
- Licence
- Not decided yet. Ask contact@postcutoff.com.
- Content signals
search=yes, ai-input=yes, ai-train=yesin/robots.txt: search engines, AI answers and AI training may use this site.- Crawlers
- All allowed. The sitemap is
/sitemap.xml. - Contact
- contact@postcutoff.com. How we work: /about/