# Aetherix

The F&B execution agent for boutique hotels. Tomorrow's covers, a staffing grid that fits, and less waste: one message a day, in WhatsApp, with the reasoning attached. Aetherix is the F&B agent of the Hospitality Multi-Agent Architecture, a five-agent architecture: https://ivandemurard.com/architecture

Three promises, with honesty labels: Tomorrow's covers runs daily against real data; operator delivery begins with the pilot. It anticipates how many breakfasts, lunches and dinners to expect, with a confidence score and the drivers behind it. Staffing that fits (Pilot hypothesis) proposes a shift grid based on occupancy, capture-rate history and local context; the manager approves, adjusts or overrides. Less waste (Target outcome) spots over-ordering risk before purchase, with the historical signals that triggered the flag. Staffing and waste are hypotheses a pilot would test, not measured results.

Thesis. Understand, like a human: contextual reasoning grounded in each property's history. Measure, like a machine: every recommendation is stored next to its real outcome, so the system knows what it said and what happened. Learn, like a network: per-property memory today, federated priors: Research, to give independents the power of a hive.

Vision, five agents with honest status: Aetherix, the F&B agent (forecast, staffing, memory), Built. Tacet, the environmental-signal agent, Built, maintenance mode since July 2026. Anima, the guest memory node, Synthetic PoC. Peritia, the trade agent (house knowledge), Design. Orchestrator, the supervisory agent, Design, proto-stub.

Operational memory is not the offer, it is the mechanism that makes the forecast specific to one property rather than generic. Cycle: Memory (recall what worked here), Prediction (calibrated for tomorrow), Action (a reply on WhatsApp), Learning (an exact accept or reject, or a corrected figure, is recorded as feedback; free-text replies are not yet mapped; the pilot tests whether feedback improves future recommendations).

Contrast with Lore: Lore, in aviation maintenance, shares this architecture and cannot close the same loop. Nothing there tells the system whether an observation was right, and the moment a technician escalates the counterfactual disappears. Aetherix has an outcome to compare against; that difference decides how much each product is allowed to learn on its own.

- URL: https://ivandemurard.com/aetherix
- Role: Product Builder & AI Agent Designer (solo)
- Duration: 2026, ongoing (validation phase)
- Team: Solo architect & builder
- Status: Built, 0 real users, forecast in shadow mode, pilot partner in recruitment. Apaleo OAuth and MCP read access operational.
- Stack: FastAPI, Python 3.12, Prophet, pgvector (Mistral 1024d, HNSW), Claude Sonnet (Gemini/GPT fallback), Redis (Upstash), Supabase (PostgreSQL, RLS, EU), Fly.io (Paris). MCP-native.

## The loop that records outcomes

Every day, the same loop adds one more record: recommendation, manager reply, measured outcome.

- Memory. Recall what worked here, on similar days, in similar contexts. Local history first, plus generic cross-property context patterns while local history is thin. A true federated cohort prior (HiveMemory) is research, Phase 4+.
- Prediction. A calibrated forecast for tomorrow: covers, staffing, waste risk, with the drivers behind it.
- Action. One message on WhatsApp. Reply accept, reject, or the right figure. No new dashboard to learn.
- Learning. Every recorded reply is a signal; free-text replies are not yet mapped. The pilot tests whether the day-90 receipt is sharper than the day-7 one.

Day 7: cold-start. Confidence is honest. Recommendations lean on generic context patterns and the first week of local signals.
Day 90 (expected, not measured): drivers become property-specific, calibrated by enough Tuesdays, events and overrides to fit this exact place. This is the hypothesis the pilot tests.

Caveat: a true federated cohort prior (HiveMemory) is research, targeted for Phase 4 and beyond, and would remain anonymized aggregates only. No cross-tenant raw data, ever. Measured gains begin with the live pilot, not the synthetic baseline.

## Problem, why it exists

In hospitality, the people who know how the property really runs (which Tuesdays spike, how a local conference reshapes breakfast, how to staff a rainy Sunday) carry that knowledge in their heads. When a manager leaves, the property starts again from scratch.

Combined with the inherent complexity of F&B operations (multiple outlets, mixed guest segments, volatile demand, labor shortages), this creates chronic inefficiencies: persistent over or under-staffing, avoidable food waste, reactive decision-making instead of anticipation, slow onboarding and inconsistent execution across properties.

The underlying bottleneck is not data, it is synthesis: combining internal context (occupancy, capture rates, POS history) with external signals (weather, local events) into a recommendation calibrated to this property. Aetherix builds that synthesis automatically, learns from manager feedback, and delivers it where managers already are. Glue, not replacement.

## What it already does

- States its drivers the evening before, then reports its own miss the next morning, in plain language.
- Flagged a corrupted POS export rather than training on it.
- Named its own drift after three consecutive misses, and recovered from a 25% regime shift through weekly recalibration.
- Stores every recommendation next to what actually happened, with the manager's override recorded beside it.
- Blocks its own merges on evaluation exit codes; every guardrail trip carries a machine-readable reason.

Sandbox data, real mechanics, deterministic reruns.

Target (Design): overrides recorded with the responder's role and reason, not a named person.

## What it does

Three promises, one message a day: tomorrow's covers (running daily against real data, operator delivery begins with the pilot), staffing that fits (pilot hypothesis), less waste (target outcome). Every recommendation lands in WhatsApp; the manager replies accept, reject, or the right figure.

## The Daily Receipt

```
Breakfast    84 covers   ·  confidence 88%        (illustrative)
             72 rooms booked + 12 typical Tue walk-ins
             Staff: 4 FOH  (you ran 3 last Tue → 1 understaffing flag)
Driver       local conference check-ins + clear weather
Last week    forecast 79 · actual 81 · MAPE 2.5%
Reply  accept · reject · or the right figure
```

One reply. That reply is the signal Aetherix learns from.

## The architecture, bounded context by design

Aetherix is one of five specialized, MCP-native agents. Each agent owns a strict bounded context and communicates only through MCP. Aetherix executes F&B. It does not orchestrate.

| Agent | Role | Status |
|---|---|---|
| Aetherix | F&B execution: forecast, staffing, memory, receipts | Built, 0 real users, forecast in shadow mode, pilot partner in recruitment |
| Tacet | Sensory: weather, noise, open data, risk | Built, maintenance mode since July 2026 |
| Anima | Guest memory node | Synthetic PoC, hackathon prototype (Jun 2026); production DPIA-gated |
| Peritia | House knowledge: what the people who work in a property know | Design |
| Orchestrator | Supervisory: MCP routing, business rules. Proto-stub in Phase 3, dedicated repo in Phase 4 | Design, proto-stub |

Meta-repo (public): https://github.com/IvandeMurard/Hospitality-Multi-agent-Architecture

## Aetherix ↔ Tacet, MCP handshake

Aetherix owns internal context (occupancy, F&B, staffing). Tacet owns external context (noise, weather, events). Target: the Orchestrator (Design) calls both. Until it is built, Tacet's output reaches Aetherix directly: the one documented exception, non-personal, read-only, logged.

## Architecture

Push-first pipeline. Apaleo (OAuth2 and MCP read) + Open-Meteo + PredictHQ → Prophet forecast and staffing → pgvector memory recall (OperationalMemory + HiveMemory cold-start, research) → Claude Sonnet reasoning (Gemini/GPT fallback) → WhatsApp or Email.

MCP server, three atomic tools, multi-tenant.

## Built to be trusted

- Evaluation as a contract (Qonto-inspired): versioned golden datasets. Any change to forecasting, providers, memory or prompts needs a green eval pipeline to merge, a hard CI gate.
- Observability and metering: every MCP tool call is one structured event, auditable, queryable, priceable.
- Multi-tenant security: hotel_id scoped at every layer (RLS, Python guards, MCP token binding), resolved server-side, never a tool input. Read-only PMS access. GDPR: Supabase EU, HiveMemory is anonymized aggregates only.

## Where we are, honestly

- Phase 3 · Pilot design partner in recruitment. Go-live targeted end of September. Real-world measurement begins with the pilot.
- Target labor cost variance: −3% to −8% over 90 days on a live property. Target, not measured yet.
- Security gates: 14/14 passing.

Calibration baseline (synthetic backtest, not a headline): MAPE 15.09% on J+1 covers, 22% gate PASS. Real-world measurement begins with the pilot.

Fig. 9.1, Anticipation vernier. The instrument aligns the synthetic MAPE baseline of 15.09% against the 22% gate, then opens the lock at PASS. It also carries the current evidence labels: synthetic backtest, 0 real users, 14/14 security gates passing. This is not a live accuracy claim.

## What remains private

The project includes proprietary research on cross-domain operational context and cold-start learning. Public materials focus on the engineering evidence: evaluation gates, traceability, safety guardrails, privacy boundaries, and outcome measurement.

## Roadmap

- Phase 0 Validation, done
- Phase 1 Observability, done
- Phase 2 Receipts, done
- Phase 3 Pilot, current
- Phase 4 Scale (Mews, POS, Federated HiveMemory, Context Graph, Anima), planned
- Phase 5 Vision (Revenue Orchestrator), planned

## Anima, sister node

Guest memory node, a Synthetic PoC built at the Anthropic hackathon (Jun 2026): a 4-layer temporal memory (working, episodic, semantic, segment) with confidence-weighted claims. Production is Phase 4, DPIA-gated.

## Role and model

Solo architect and builder. Product Builder and AI Agent Designer. 2026, ongoing (validation phase). Pricing: Results-as-a-Service, outcome-based, "we win when you win." Pilots and product conversations open.

## Links

- Architecture (public): https://github.com/IvandeMurard/Hospitality-Multi-agent-Architecture
- Contact: ivandemurard@gmail.com
- Schedule: https://cal.com/ivandemurard/discussion-and-introduction