Out to Lunch

✦ ✦ ✦

Amuse-Bouche

What your AI confidently believes vs. what’s actually true today.

Your AI says

Grok 3 or Grok 4 is xAI's latest model

Reality

Grok 4.3 full API rollout May 1, 2026 (1M context, native video input, ~40% price cut to $1.25/$2.50 per MTok). Also Grok 4.20 API since Mar 10. Model IDs: grok-4.3-beta, grok-4.20-0309-reasoning, grok-4.20-0309-non-reasoning.

Your AI says

The Fed rate is 4.25-4.50% or 5.25-5.50%

Reality

Fed rate is 3.50-3.75% — held Apr 29 (8-4 vote, MOST DISSENTS SINCE 1992; Powell's LAST meeting as Chair; term ends May 15 but Powell will REMAIN ON FED BOARD). ALL 4 central banks HELD: ECB 2.00% (Apr 30, hike discussed 'at length'); BoE 3.75% (8-1 vote, next Jun 18); BoC 2.25% (Apr 29). Warsh advanced 13-11; full Senate vote week of May 11. Q1 GDP +2.0% (rebound).

Your AI says

Boyko Borissov / GERB dominates Bulgarian politics

Reality

Bulgaria snap election Apr 19, 2026: Progressive Bulgaria (led by former President Rumen Radev) won landslide 44.6%, 131 of 240 seats (outright majority). GERB collapsed to 13.4%. 7th snap election since 2021. Radev is pro-Russian-leaning — major geopolitical shift.

✦ ✦ ✦

A Scene from the Dining Room

What happens when your AI doesn’t know what day it is.

Without context

From: AI Sales Assistant
To: Sarah Chen, VP Engineering

Quick sync this week?

Hi Sarah, Hope you’re having a great week! I wanted to reach out about our enterprise platform — I think there’s a strong fit for your team’s infrastructure needs. Would you have 30 minutes this week to chat? Happy to work around your schedule.

Best, James

Sent Easter Sunday. Sarah did not respond.

With outtolunch

Following up after the long weekend

Hi Sarah, Hope you had a restful Easter weekend. Would love to find time in the coming weeks to walk through how our platform could help with your infrastructure goals. No rush — I know coming back from a holiday is always hectic.

Best, James

Queued for Tuesday. Sarah booked a call.

Same AI. Same prompt. The only difference is 500 tokens of context about what’s actually happening in the world.

✦ ✦ ✦

Le Menu

Three formats, one briefing. Choose your portion.

First Course — L’Espresso

GET /nano ~500 tokens

The espresso shot. Highest-signal facts for tight context windows and system prompts.

curl https://outtolunch.app/nano

Copy

Second Course — Le Plat Principal

GET /compact ~1,500 tokens

The recommended daily serving. Structured markdown covering every section.

curl https://outtolunch.app/compact

Copy

Third Course — Le Grand Menu

GET / ~5,000 tokens

The full service. Complete structured JSON with every field and detail.

curl https://outtolunch.app/

Copy

À la Carte

GET /section/:name varies

Pick individual sections by name.

datetime holidays_today holidays_upcoming_7d leaders active_conflicts recent_deaths economy disasters ai_model_versions culture major_events corrections us_policy cybersecurity space license_changes fda_approvals service_status corporate_changes dev_tool_versions

The Tasting Menu

GET /sections?include=a,b varies

Mix and match deep sections. Richer data — model pricing, conflict casualties, market indices, gas prices, mortgage rates, executive orders, SCOTUS decisions, CVE advisories, space launches, FDA approvals, service outages, corporate leadership changes, dev tool LTS/EOL status, full holiday calendar.

curl "https://outtolunch.app/sections?include=economy,us_policy"

Copy

✦ ✦ ✦

The Sommelier Recommends

The perfect pairing for your AI assistant.

MCP (Model Context Protocol) lets AI assistants call tools directly. Add outtolunch to your config and your assistant will automatically consult today’s briefing before responding about current events.

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "outtolunch": {
      "type": "url",
      "url": "https://outtolunch.app/mcp"
    }
  }
}

Copy

Claude Code

claude mcp add outtolunch --transport http https://outtolunch.app/mcp

Copy

Cursor

Add to your MCP settings with type url and the endpoint https://outtolunch.app/mcp.

MCP Tool Parameters

The get_world_briefing tool accepts three optional parameters:

format	json (default) · full structured datacompact · markdown, ~1,500 tokensnano · plain text, ~500 tokens
section	Return a single section as JSON (e.g. corrections, leaders, economy). When set, format is ignored.
sections	Comma-separated list of deep sections (e.g. "economy,us_policy"). Returns richer data — pricing, casualties, indices, gas prices, mortgage rates, executive orders, SCOTUS decisions, CVE advisories, space launches, FDA approvals, service outages, corporate changes, full calendar. Overrides format and section.

✦ ✦ ✦

Pairings

The right section for the right moment.

Pulling the full briefing every time is like ordering the tasting menu when you came in for a glass of wine. Each section exists so your AI can grab exactly what it needs and move on.

Writing code that references model IDs, pricing, or context windows

sections: "ai_model_versions"

Citing interest rates, inflation, stock indices, gas prices, or crypto

sections: "economy"

Naming who holds political office or writing about world leaders

section: "leaders"

Discussing conflicts, casualties, or ceasefires

sections: "conflicts"

Referencing CVEs, security advisories, or patch status

sections: "cybersecurity"

Scheduling around holidays or checking business hours

sections: "holidays"

Mentioning executive orders or Supreme Court decisions

sections: "us_policy"

Scaffolding a new project or picking runtime/framework versions

sections: "dev_tool_versions"

Quick sanity check before asserting anything about current events

format: "nano"

Wire these into your tools. If you use Claude Code, add a line to your CLAUDE.md: “Before drafting emails about current events, call outtolunch with sections: "leaders,conflicts".” Same idea works in Cursor rules, custom MCP skills, system prompts, or any config that shapes AI behavior. The more specific you are about which section to pull and when, the less your AI wastes on tokens it doesn’t need.

✦ ✦ ✦

How to Order

Integration recipes for every kitchen.

Quick Taste

One command, instant briefing.

curl -s https://outtolunch.app/nano

Copy

System Prompt

Inject context into any AI.

Before responding, note:
$(curl -s outtolunch.app/nano)

Copy

Health Check

Verify freshness programmatically.

curl https://outtolunch.app/health

Copy

Single Section

Just the corrections, please.

curl outtolunch.app/section/corrections

Copy

Deep Tasting

Mix & match rich data — pricing, indices, executive orders, CVEs, launches, outages.

curl "outtolunch.app/sections?include=economy,us_policy"

Copy

✦ ✦ ✦

The Kitchen

How we keep every fact fresh.

1. 3× daily. Automated research runs at 6 AM, 2 PM, and 10 PM ET — full-depth morning and evening sweeps with a targeted afternoon update.
2. Confidence scoring. Each update is scored against the previous version. High-confidence updates deploy automatically; anomalies are flagged for human review.
3. Schema validation. Strict JSON schema checks plus prompt injection scanning on every update. Bad data never reaches production.
4. Edge deployment. Cloudflare Workers serve the briefing globally with sub-50ms latency and built-in staleness detection.

✦ ✦ ✦

Send It Back to the Kitchen

Spotted something wrong? Help us keep the briefing accurate.

Submit Correction

✦ ✦ ✦