All decks
Anaïs AI (working app)Pharoah Technology · deck · narration · production guide · generated 9 September 2026
02What it is

The facts, from the register.

ProductAnaïs AI (working app)
Register keysol-ai-app
OwnerPharoah Technology Ltd (Craig Attoh and Phil Pereira, 50/50)
Live atanaisai.co.uk · HTTP 200 on 9 September 2026
Price£29 a month · [object Object]
Built from4 register parts, 4 live, 2 probe-verified
Listed14 August 2026

What it does

The working Anaïs AI app — a council chat that routes each prompt across models through the shared gateway.

Runs on

Expense Management
Sol AI working app (TanStack Start council chat)
Meeting Note Taker — universal transcript→notes engine
AI model router (LiteLLM ai-proxy-gateway)
Source: estate register (products, product_parts, builder_blocks) · deploy probe 9 September 2026Evidence live
03The frameWhat this buyer's profession checks

We will measure it against the buyer's own change-control.

Three checks a buyer in this field applies, whether or not they name the standard. We use them as the frame for everything that follows.

Check 1

A control between the work and the submission

Check 2

A process fit for this buyer, not a generic one

Check 3

Readiness for the next rule change

Frame chosen by BRG as a summary of the buyer's own change-control. Verify the wording with the body before external use.No accreditation is claimed
04Why it's better · 1 of 6Check 2 · fit for purpose

The job as the register describes it.

Picking a model per task is guesswork, so people default to the most expensive one for everything.

Who carries it

People who need AI assistance and struggle to choose the right model for each task.

Source: register pain_point · audienceThe problem, stated plainly
05Why it's better · 2 of 6Check 2 · fit for purpose

What it does instead.

Multiple AI minds confer on every question, then synthesize one unified answer. Smarter than any single model.

The working Anaïs AI app — a council chat that routes each prompt across models through the shared gateway.

The parts doing the work

Expense Management

Universal expense capability: claim capture (portal/WhatsApp), multi-line receipts, approval workflow, and reimbursement intents on the forge.invoicin

Sol AI working app (TanStack Start council chat) verified

Sol AI working app: TanStack Start (React 19 SSR + server APIs) + Supabase (auth/Postgres) council-chat. Streaming multi-agent chat that synthesises o

Meeting Note Taker — universal transcript→notes engine

ONE reusable engine closing the largest gap cluster (53 products needed "meeting note taker"). Reuses rail.llm-gateway; per-product behaviour = a conf

AI model router (LiteLLM ai-proxy-gateway) verified

LiteLLM AI model router with pinned free/core aliases, cost-aware fallback routing, live health/model inventory, and live gateway chat proof. Direct f

Source: register purpose, public_blurb, product_parts joined to builder_blocks4 of 4 parts shown
06Why it's better · 3 of 6Check 1 · controls

Proof it exists, not a roadmap.

200

HTTP status of anaisai.co.uk on 9 September 2026 · real-content-verified

2/4

register parts verified by probe

4

parts wired live

HTTP 200

functional surface probe

Open anaisai.co.uk during the talk. Anything the deck claims should be visible there.

Source: register deploy_probe_code, deploy_probed_at, product_parts.verified_by_probeProbed, not promised
07Why it's better · 4 of 6Against the market, honestly

Where we lead, and where we don't.

Category: multi-model AI comparison and answer assurance. Nearest rivals: Poe, ChatHub, Perplexity, OpenRouter Chat, Chatbot Arena; strongest is Poe on breadth. Our edge: rivals show several answers side by side and leave you to read them; we surface DISAGREEMENT as the signal — where the models diverge is where the factual risk sits, which is the question a reviewer actually needs answered. How we beat them: comparison UIs sell access to many models; we sell a verdict on reliability, which is what someone shipping an AI feature is buying.

Not independently sourced This verdict was written from an analysis whose only source was the model itself, not scraped competitors (analyst gtm-loop-worker). Treat the competitor names and positions as a working hypothesis until the market research is re-run; do not quote them to a buyer.
Source: register vs_competition (18 August 2026)Named competitors are the register's, not the presenter's
08Why it's better · 5 of 6Check 2 · fit for purpose

Built for one kind of buyer.

People who need AI assistance and struggle to choose the right model for each task.

Source: register audience, serves_nichesOne buyer, not everyone
09Why it's better · 6 of 6Price is part of the scope

What it costs, in one line.

£29 a month

Basis: [object Object]

Buying route

Payment on enquiry
software
Scope named on the next slides, so the price has a boundary
Source: register price_gbp, billing_interval, pricing_basis, payment_link_activeNo hidden rate card
10What's inside

The parts, with their verification state.

PartCategoryStateVerified
Expense Managementfull-stacklive
Sol AI working app (TanStack Start council chat)uilive_wired
Meeting Note Taker — universal transcript→notes enginebackendlive
AI model router (LiteLLM ai-proxy-gateway)integrationlive_wired
Source: product_parts joined to builder_blocks · 4 of 4 shown probe-verified   not yet
12Why packaging matters

Five things a package does that a price cannot.

1
Scope becomes visible. The parts are named, so nothing is assumed.
2
Boundaries are explicit. What is outside is written down before the first invoice.
3
It is auditable. A package is a documented process, which is what the buyer's own change-control looks for.
4
Buyers compare like for like. A referral needs a defined thing to refer.
5
Switching fear drops. The move is inside the package, not a project bolted on the front.
Source: BRG packaging doctrine; frame from slide 3Packaging = a compliance instrument
13This packageCut by what each part does

Anaïs AI (working app), as a package.

full-stack

Expense Management

Universal expense capability: claim capture (portal/WhatsApp), multi-line receipts, approval workflow, and rei

live
ui

Sol AI working app (TanStack Start council chat)

Sol AI working app: TanStack Start (React 19 SSR + server APIs) + Supabase (auth/Postgres) council-chat. Strea

verified
backend

Meeting Note Taker — universal transcript→notes engine

ONE reusable engine closing the largest gap cluster (53 products needed "meeting note taker"). Reuses rail.llm

live
integration

AI model router (LiteLLM ai-proxy-gateway)

LiteLLM AI model router with pinned free/core aliases, cost-aware fallback routing, live health/model inventor

verified
Source: product_parts grouped by builder_blocks.category4 groups
14Anatomy of the packageOn one page · 3 items of work left, filed

A package reads like a control document.

Inside the scope

  • Sol AI working app (TanStack Start council chat)
  • AI model router (LiteLLM ai-proxy-gateway)

Evidenced, and how

  • Live URL probed: HTTP 200 on 9 September 2026
  • 2 parts probe-verified, latest 16 July 2026
  • Functional surface: HTTP 200
  • Register row updated with every probe

Outside the scope today · work left

  • No build gap recorded in the register
  • 2 parts still to verify by probe
  • 1 register field to fill (vs_competition_unsourced)

Who signs, and where the work sits

  • Pharoah Technology Ltd owns the product and the data layer
  • 0 build packets proposed · 3 management items pending
  • Change log kept in the register
Source: product_parts (inside), deploy and part probes (evidenced), v_deck_work_left (work left; filed to manager_work and build_packets by deck_work_left_sync)One page
15Across the family

Where it hands over.

Anaïs AIgateway.pharoahtechnology.com
Anaïs AI (LLM router / reseller)anaisai.co.uk
ATS Résumé Scanner (free)ats.pharoahtechnology.com
Pharoah TechnologyAnaïs AI (working app)anaisai.co.uk
Aurora AIauroraai.pharoahtechnology.com
Aurora Guardauroraguard.co.uk
BrandForge Probrandforge.pharoahtechnology.com
Source: register rows sharing client_key pharoahOne house, many doors
1 / 16
Arrow keys or space · N narration · P print
Slide 1Narration Words Running total

Production guide

How to turn these sixteen slides into a presentation video. Same structure as every deck in the family, so a crew that has made one can make them all.

1 · Deliverable

ItemSpec
Format16:9, 1920×1080, 25 fps, H.264 MP4 plus WebM. Burned captions and a sidecar SRT.
RuntimeNarration below plus screen captures of the live product; typically 8 to 12 minutes.
Framethe buyer's own change-control. Verify the wording of the three checks with the body before external use.
BrandPharoah Technology family palette as rendered here. Green only for verified and live states; red only for exclusions.
VoiceCalm, exact, never theatrical. Read figures and dates in full. Numbers are the register's, and the register is cited on every slide.

2 · Runtime plan

#SlideNarrationWordsStarts

3 · Screen captures

4 · Full narration

5 · What this deck must not say

  • No numbers that are not on a slide with a register source.
  • No accreditation by the buyer's own change-control; the frame is a lens, not a badge.
  • No named competitor read aloud unless the audience asks; the register's names are on slide 7 for the presenter.
  • No client names, volumes or tenure unless the register carries them.

6 · Regenerate

This deck is generated from the estate register. Update the product row, re-run the deck build, redeploy. Do not hand-edit this file.