# ai-sota-feed-bot — llms.txt This file helps LLM agents understand and navigate this site. ## What this site is - AI SOTA feed focused on agent engineering, LLM platform updates, releases, and selected research. - The homepage is a reader UI backed by JSON APIs. - Primary production URL: https://ai-sota-feed-bot.vercel.app/ ## Fast navigation map - Home UI: / - JSON feed (main): /api/feed - RSS feed: /api/rss - Telemetry ingest: /api/events (POST) - Client runtime config: /api/client-config ## Recommended access path for LLMs 1) Fetch `/api/feed?limit=200` to get current merged feed items. 2) Read `runs` metadata to understand recency and run history. 3) Use item fields (`source`, `published`, `v2_slot`, `llm_category`, `v2_final_score`, `labels`) for filtering/summarization. 4) If asking for older content, set explicit date range or use larger window: - `/api/feed?from=&to=&limit=500` 5) If you need strict source/topic slices, use labels: - `/api/feed?label=frontier_official` - `/api/feed?label=platform&label=frontier_official` ## Feed API quick reference Endpoint: `GET /api/feed` Query params: - `from` (ISO datetime, optional) - `to` (ISO datetime, optional) - `limit` (1..500, default 200) - `label` (repeatable) - `labels` (CSV alternative) - `blend_tier1` (`1` default, `0` to disable) - Tier-1 blend knobs (optional): - `tier1_fresh_cap` - `tier1_insert_after` - `tier1_min_quick_score` - `tier1_max_per_source` Response highlights: - `mode`: `history` or `latest` - `runs[]`: run timestamps and item counts - `items[]`: feed entries with ranking and metadata - `available_labels[]`: label counts - `tier1_blend`: blend diagnostics - `personalization`: optional ranking adjustments by anon user id ## Common LLM tasks - "Top frontier lab updates today": `/api/feed?label=frontier_official&from=&limit=100` - "Only platform items": `/api/feed?label=platform&limit=200` - "Recent OpenAI/Anthropic coverage": fetch `/api/feed`, then filter `source` in `{openai_blog, anthropic_newsroom, anthropic_engineering, claude_blog}`. ## Caveats - Homepage defaults to a recent time window preset. Older items may exist in API but not be visible until window is widened. - Ranking is slot-constrained for diversity; not every valid source item appears in final Tier-0 output. ## For deeper instructions - See: `/llm-guide.txt`