Six agents · One council · Always watching

While you run the floor,
six agents run your margin.

The Watch reads your POS. Cost Sentinel reads your invoices. Menu Engineer reads your mix. Comp/Void Coach reads your void slips. Pour Coach watches the bar. Schedule Critic watches the draft. The Analyst frames every find in operator voice. The Coach stages a one-tap play. You commit it in Slack — or you don't.

Book a 15-min walkthrough Meet the council
All 6 running on real Toast + R365 + 7shifts data 71% batting avg · last 30d Built by an operator, for operators
app.heyrai.io / agent-activity / live All 6 agents running
07:01brief sentmorning brief delivered → #warren-gm, #providence-gm, #newport-gm11 plays
06:14comp/voidWarren · J. Reyes 47% of void $ last 14d · 22 voids, last-90-min cluster, "wrong order" tag$864
06:12cost sentinelNewport · beef short rib +18% w/w on 4/30 invoice · vendor: NE Provisions−$340/wk
06:08menu engineerProvidence · burrata salad mix −22% over 21d · margin holding · drift, not dogmenu drift
06:02the watchWarren · drink attach 31% vs 38% target · 4 consecutive operating days · severity high−$420
05:58pour coachNewport · house red → reserve flight conversion at +$3.20/cover on weekend dinners+$1,150/wk
05:55schedule criticWarren · Sun brunch labor 34% vs 26% target on next week's draft (cut 1 host, shorten line cook)−$420
04 detector pass · 4 flags raised
05:45analystframed 4 flags · pulled correlations: shift_window, reason_tags, peer_baseline, vendor_cohort4 framed
05:42coachstaged 11 plays · 7 GM-audience, 3 owner-audience, 1 escalation · all freshness-checked11 staged
02 overnight detector run · 14d window · 3 locations
02:14orchestratorcouncil pass complete · 6 detectors · 3 locations · total 1.9s · status okhealthy
6 agents· 3 locations· last run 06:14 ET· grading active· time-of-day hold 23:00 → 07:30 ET
The council · Six specialists, two framers

Each agent watches one thing and watches it well.

No general-purpose AI. Every agent has a single job, a single data source, and a single output: a flag with receipts. The Analyst frames it. The Coach stages a play. Below: what each one is doing today on Chomp data.

TW
The Watch
Watches · POS variance
Running
Reads every check, every void, every comp out of Toast. Compares the day's drink attach, starter attach, PPA, avg check, and void rate to your targets — not an industry average. Severity grades by aging.
Today's find · Warren
drink attach 31% vs 38% target · 4 days aged · severity high
CS
Cost Sentinel
Watches · Invoice spikes
Running
Ingests every R365 AP invoice the moment it lands. Detects price spikes per SKU per vendor, vendor-consolidation opportunities, and true food cost % drift — independent of POS, runs portfolio-wide.
Today's find · Newport
beef short rib +18% w/w · vendor: NE Provisions · backup quote available
Menu Engineer
Watches · Mix & margin
Running
Joins Toast item mix to R365 cost data. Surfaces menu drift (mix slipping but margin healthy), menu dogs (slipping with margin pressure), price-up candidates, and high-margin items losing share to low-margin ones.
Today's find · Providence
burrata salad mix −22% over 21d · margin holding · drift, not dog
CV
Comp/Void Coach New
Watches · Void slips & comp pad
Running
The Watch flags the rate. The Coach attributes it — to a server, a manager-PIN, a daypart, a reason tag — and stages a 5-minute walk-through play the GM can run before the next shift.
Today's find · Warren
J. Reyes 47% of void $ ($864) · 14 of 22 in last 90min of shift · "wrong order" 8× peer rate
PC
Pour Coach
Watches · Beverage trade-up
Running
Reads pour-by-pour bar data. Identifies trade-up opportunities (house red → reserve flight, well vodka → premium) by daypart, server, and guest cohort. Quantifies the per-cover lift before staging.
Today's find · Newport
house red → reserve flight at +$3.20/cover on weekend dinners · +$1,150/wk
SC
Schedule Critic
Watches · Labor drafts in 7shifts
Running
Grades next week's draft against forecast covers, role-mix history, your daily targets, and OT risk. Flags the daypart-shifts with the most $ at risk — before publish, not after the breach.
Today's find · Warren
Sun brunch labor 34% vs 26% target on next draft · cut 1 host, shorten line cook 1.5h
A
The Analyst
Reads every flag the council raises. Pulls correlations (shift window, reason tag, vendor cohort, peer baseline) and writes a one-paragraph frame in operator voice. No LLM hallucination — it's template-stitched from real data.
C
The Coach
Reads framed flags. Pulls a play from the playbook (audience-routed: GM, owner, you), checks freshness (don't alert if KPI already recovered), and stages it as a one-tap commit in Slack — with a 14-day verdict baked in.
A flag in flight · Comp/Void Coach · Warren · this morning

One real flag, five stages, ninety seconds.

From overnight detection to a play in the GM's Slack to a 14-day verdict. This is what the council did at 6:14 a.m. on real Warren data.

1
Detect
Concentration crossed 40%
Comp/Void Coach pulls 14 days of voids by server. One server is 47% of dollars against a peer median of $86. Threshold tripped.
voids · 22
$ · $864
peer median · $86
share · 47%
2
Frame
Why it matters, in your voice
"J. Reyes ran 22 of the 47 voids last 14 days — $864 against a peer median of $86. 14 of 22 cluster in the last 90 min of his shift. Reasons skew 'wrong order' (8 vs peer 2)."
correlations checked · 4
hits · shift_window
+ reason_tags
framed at · 05:45 ET
3
Stage
A 5-minute walk-through
Pull J. Reyes' last 14d void slips before his next shift. Walk the closing-rush procedure. Flag the "wrong order" tag specifically. Set 7-day target: voids ≤ $40/shift.
audience · gm
expected recovery · $540
window · 7 days
alternates · 2
4
Slack
Lands in #warren-gm
Real-time commit. Freshness-checked, time-of-day held until 7:30 a.m. ET, deduplicated against open flags.
Rai App 6:14 AM
Warren — voids 3.8× normal last 14d ($1,840). One server, J. Reyes, accounts for $864 across 22 voids — most in the last 90 min. Worth a 5-min walk before his next close.
Commit play
5
Verdict
Self-graded at 14 days
If the GM commits, the Coach watches the next 14 days. Did J. Reyes' voids/shift drop under $40? Did the location's total voids drop? Hit, miss, or partial — recorded.
outcome window · 14d
target · ≤$40/shift
last verdict · hit
$ saved · $510
The 24-hour loop · A typical Warren weekday

Detection at 2 a.m. Brief at 7. Plays through the day.

The council runs against your data overnight. Plays land in Slack and email by the time you walk in. Real-time alerts fire as variances appear during ops. The line goes quiet at 11 — no late-night pings.

02:14 Detector pass 6 specialists · 14d window · 1.9s total
05:45 Analyst frames 4 flags · operator voice
05:55 Coach stages 11 plays · audience-routed
07:00 Morning brief Slack + email · per-location card
11:32 Real-time alert lunch rush · prep variance
16:05 Real-time alert dinner setup · drink attach soft
20:18 Real-time alert comp pad spike · post-9pm
23:00 Hold begins no Slack until 07:30 ET
71%
Batting avg
last 30d graded by agent_outcomes
$14.2k
Identified · 30d
3 locations Toast + R365 + 7shifts
47
Plays committed
1-tap from Slack last 30d
4h 22m
Avg flag → action
flag staged → GM commit last 30d
Already wired into the stack you run
Why it works

Not another dashboard. A council with receipts.

Every flag is graded. Every play is auditable. Every recommendation has provenance. We tell you when we miss.

Specialists, not generalists

One agent, one job, one data source.

An agent that watches voids isn't also writing schedules. Specialists stay sharp. The council pattern means a new specialist (like Comp/Void Coach) plugs in without touching anything else.

Vs. dashboards: dashboards make you look. Agents stage the play.
Frames in operator voice

Template-stitched. No LLM hallucination.

The Analyst pulls correlations from real data and stitches a frame in your voice — no model drift, no invented numbers. Every chip in the frame links back to the row that produced it.

Vs. AI summaries: summaries are vibes. Frames are receipts.
Self-grading

14-day verdicts. Hit, miss, or partial.

Every committed play is checked at 14 days. Did the metric move? Did dollars come back? The batting average is published — you see what worked, what didn't, and why.

Vs. "AI for restaurants": we publish our miss rate.
Pricing · per location, per month

Three tiers, same nervous system.

Three tiers that match how you want the loop to work for you — start with the generalists, layer in specialists, or co-build a custom council. Annual billing; quarterly is +15%. Setup, calibration, and POS / scheduling wiring are included in every tier.

01 · Floor
The Starter.
$299 / location · month
The three generalists running on your real data. Best if you want to feel the loop on a single concept before wiring the specialists.
  • The Watch, Analyst, Coach — the three generalists in the loop, daily.
  • POS + scheduling app wired (live sync through ops hours, multi-month backfill).
  • Plan HQ with target picker (Floor / Optimal / Stretch), trend pills, daily-target outcomes.
  • Loop Brief emailed every morning · the 3-line headline of what changed.
  • Self-grading: every flag carries its receipts.
Start with the Floor →
Recommended
02 · Optimal
The Operator.
$599 / location · month
The full live loop. What we run in production today across multi-unit operators — every shipped specialist, accounting invoices, GM Slack routing.
  • Everything in Starter, plus the specialists below.
  • Pour Coach — surfaces five-figure weekly trade-up opportunity in live runs.
  • Schedule Critic — Saturday 11pm verdict with role · hours · shifts.
  • Cost Sentinel trio — Price Spike, Vendor Consolidation, True Food Cost%.
  • Comp/Void Coach — server-attributed void and comp coaching, with receipts.
  • Accounting invoices ingested + plays auto-routed into your GM Slack channels.
  • New specialists added to the roster as we ship them — no extra cost while you're a customer.
Talk to Sam →
03 · Custom
The Studio.
$1,200 / location · month
Built for operators who want a co-builder, not a vendor. Custom specialists scoped to your concept. Limited to a handful of partners.
  • Everything in Operator.
  • Custom specialist agents — 1 new agent built to your job-to-be-done per quarter.
  • Quarterly ops review with the founder · numbers, agent batting avg, what to add.
  • Outside-system feeds — reservations, reputation, vendor portals, beverage distributors.
  • Dedicated #loop-ops Slack · same-day response during weekday ops hours.
  • Volume pricing past 5 locations.
Build with us →
The 5× pact

your monthly fee in named recoverable cash within 30 days — or your next month is on us.

If the council doesn't surface 5× the Optimal tier fee in named opportunity (item, dollar value, named owner) inside your first 30 days, your second month is free. We track identified and captured per location, with receipts on every flag — the Brief shows you both numbers daily.

Setup and calibration included · cancel anytime · 30-day side-by-side trial against your current dashboards.
What we don't promise: that captured equals identified. The agents surface the opportunities — the captured number depends on the GM committing the play and the team following it on the floor. We grade ourselves on both.

How Rai onboards

Loop live in a couple of weeks.
Roster keeps growing after.

No app to install, no IT project. We wire the loop into your stack, calibrate it to how you run, and ship new specialists into the loop as we build them — included while you're a customer.

1

Wire the senses

Share read-only credentials for your POS, scheduling app, and accounting software. We handle the OAuth, the section mapping, and the multi-month backfill so the loop has baselines from day one.

week 1
2

Calibrate the loop

We sit with you on operating days, section names, hard-floor targets, role mappings, and concept tells — so every agent grades against your normal, not an industry average. Stretch targets seeded from your trailing 56d.

week 1–2
3

Loop goes live + the roster grows

The Watch, Analyst, and Coach start staging plays. Pour Coach, Schedule Critic, Cost Sentinel, and Comp/Void Coach switch on next. New specialists ship into the loop as we build them — often shaped by questions your team raises during ops review.

week 2 → ongoing
From the builder
"We build small, accountable agents — one specialist, one job. Beverage trade-up. Schedule critic. Comp/Void Coach. The point isn't that any single agent is brilliant. It's that operators can't watch everything, and a loop of small accountable specialists can. Every commit, override, and outcome our team logs makes the next agent sharper than the last one we shipped."
Sam Glynn · Principal, Water Street Advisors · Founder, Rai

The agents flag it.
You commit it. The dollars are real.

Private beta, personally configured by an operator. 20 minutes on a screenshare is usually enough to know if a loop fits how you run. We'll show you a live Plan HQ on a working operator's data — the wins, the misses, and the batting average.

30-day side-by-side trial Setup included Cancel anytime