Pricing

Boutique first — Enterprise when the office grows.

Hero SKU: Boutique Team — up to 5 seats for £449 / month plus £299 onboarding. If your team already uses ChatGPT for client admin, start with Boutique (£449) — ChatGPT speed. Private UK node. Named unlock before anything client-facing leaves. Book a 45-min desk demo; paid pilot £299 credits to setup.

One leaked client file usually costs more than Boutique Team (£449 / month). A2Zai is the private desk that keeps the speed and removes the paste-into-ChatGPT habit.

Regional Practice: £521–£1,728 / month (6–24 × £72) + £499 setup. Enterprise Operations: £1,800–£3,600 / month (25–50 × £72) + £799 setup. Paid desk pilot £299 credits to Boutique setup within 30 days.

Paid pilot

Paid desk pilot

One desk · up to 5 days

£299

one-off · credited to Boutique setup (£299) if you convert within 30 days

Ring-fenced UK node on your own messy inbox samples
Not a free forever trial — £299 reserved against Boutique onboarding
Invitation 5-day pilots remain for hot deals only

Request the £299 paid pilot

Hero SKU

Boutique Team

Up to 5 seats · hero SKU

£449

flat / month for up to 5 seats · £299 onboarding & private node (one-off)

Autonomous intake structuring and draft desk on your private UK node
Admin Review / human unlock where you configure it
Fair-queue inference · support Mon–Fri 09:00–17:30 UK

Start Boutique

Subscription

Regional Practice

6–24 seats · £72 × seats

£521–£1,728

£72/seat / month · 6–24 seats · £499 setup

Everything in Boutique Team
Pay only for the seats you need between 6 and 24
Same private-node queue, metering, and privacy defaults

Start Regional

Subscription

Enterprise Operations

25–50 seats · £72 × seats

£1,800–£3,600

£72/seat / month · 25–50 seats · £799 setup · fits one cloud GPU node

Autonomous multi-step planning with self-checks before delivery
Map-reduce for long / multi-doc files on a private UK node
Dual-lane Admin Review for client-facing drafts
Dual-layer firm memory gated by your privacy settings
25–50 named seats · metering · fair-queue GPU protection

Start Enterprise

How seats work

Published bands — setup plus monthly

Book a 45-min desk demo → choose Boutique (hero), Regional, or Enterprise → pick named seats → pay the one-off onboarding & private node fee plus the monthly subscription on Stripe → create the organisation admin login → invite colleagues up to your seat count. Optional £299 paid pilot credits to Boutique setup within 30 days. Invitation 5-day pilots are for hot deals only. A separate Implementation & Security Audit is optional and quoted only when scoped.

Whitelisting your firm’s office IP addresses so outside networks cannot reach the portal.
Formatting and securely parsing your initial internal document templates.
A short security walkthrough of Sovereign Data Protection on Staffordshire hardware.

Commercial scope

What the monthly fee includes

Seats are named people inside your band. Inference uses a fair queue (FIFO lanes + round-robin) — peak dumps wait their turn. Output is an administrative draft until a responsible human in your organisation signs it off.

Autonomous multi-step planning (plan → act → observe → validate) with bounded replan when outcomes miss the goal.
VRAM-safe map-reduce for large and multi-document files — chunk, extract, consolidate, then hand a clean summary to the main agent.
Parallel tool calling and background agent jobs so long work does not lock the desk UI.
Dual-lane routing: autonomous for internal categorisation/extraction; Admin Review for client-facing drafts and issue unlock.
Inbound lead categorisation, contact extraction, and routing flags from forms and email.
Draft desk: non-binding administrative acknowledgement templates for clerks to edit and send (A2Zai does not transmit mail).
Dual-layer memory — episodic session state plus firm long-term preferences/summaries when durable retention is enabled.
Isolated multi-tenant node architecture — one organisation’s matter text does not appear in another’s drafts.
Per-tenant token / request metering with configurable soft and hard caps to protect GPU availability.
Fair queue (FIFO lanes + round-robin) with explicit queue position — peak dumps wait their turn; no consumer-chat latency promise under full load.
Inference on Staffordshire hardware — client text is not used to train public AI models. Ephemeral matters by default.
Horizontal scale: add private GPU serve nodes when you outgrow one card; training hardware stays separate.
De-ID / Stay-on-Page where configured; human unlock where Admin Review is enabled.
Hard context ceiling (default 8,000 tokens; raise on larger GPUs) so concurrent seats stay inside the host VRAM envelope.

Explicitly not included

Never claim Legal AI, LegalTech, law-firm AI, solicitor software, regulated professional advice (legal, tax, medical, or otherwise), unsupervised advice, settlement figures, or liability admissions.
Any right to issue AI output to a client or third party without Admin Review unlock (professional-office) or named human review and sign-off (other desks).
Auto-sent contextual AI SMS (template acknowledgements only where SMS is enabled).
Guaranteed instant generation times, or that every draft is correct without named human review.
A promise that every concurrent long job finishes at consumer-chat latency — seats share the fair queue (FIFO lanes + round-robin).
Dedicated GPU reservation or priority-queue SKUs (roadmap — not provisioned on this founding node).
A claim that client text never touches any third-party processor (Cloudflare Access and Stripe are used for edge login and payments).
Full NER anonymisation, unsupervised complete contracts, or native 128k research context (product context is capped for reliable local inference).
Per-organisation model fine-tuning from uploaded templates (style/structure brief only).

Optional

Extras — quoted separately

Keep the base seats clean. Grow capacity or compliance only when the organisation asks for it.

Extra seats beyond your paid headcount
Additional named seats on the same organisation node, quoted as an upgrade to the next published seat band.
Persist-to-disk drafts
Optional on-disk draft retention when your compliance policy requires it (off by default).
On-site walkthrough day
Optional full day at your office if managers want a guided tour — not required; the Jobs UI is built so anyone can start without a course.
Dedicated GPU / priority scheduling (roadmap)
Quoted only when we can provision reserved capacity on the Staffordshire node. Not available on the shared founding fair queue today.

England & Wales schedule

Service Licence clauses

Paste into the order form / schedule. Written for administrative human-in-the-loop use, UK GDPR, fair queue (FIFO lanes + round-robin) performance, and exclusive English jurisdiction — settle final wording with your counsel before live charging.

Administrative scope & human-in-the-loop

The Service is an automated administrative intake, routing, and drafting assistant for organisations in any industry. It does not provide regulated professional advice (including legal advice under the Legal Services Act 2007). The Customer must ensure a responsible human reviews, amends where required, and signs off 100% of Service output before it is issued to any client or third party. Nothing in the Agreement excludes or limits liability for death or personal injury caused by negligence, fraud, or any other liability that cannot lawfully be excluded under English law; subject to that, the Provider does not accept liability for professional advice decisions taken by the Customer.

UK GDPR & data privacy

The Provider operates the Service in accordance with UK GDPR and the Data Protection Act 2018. Customer content is processed for inference on United Kingdom hardware (Staffordshire) and is not used to train public or Vendor foundation models. Binding zero-data-retention for training applies. Edge authentication and card payments transit named processors (Cloudflare Access; Stripe) as disclosed to the Customer — inference is not performed by those processors. Default draft mode is ephemeral (session/RAM); durable disk retention is optional. Operational logs required for security, audit, and support are limited to what is necessary to run the Service for that Customer.

Fair queue (FIFO lanes + round-robin) & support hours

Inference uses a fair queue (FIFO lanes + round-robin): work is ordered within size lanes and scheduled fairly across seats so no single user monopolises the GPU. The Provider does not warrant instant text generation or fixed per-request latency. Large simultaneous dumps (for example many long documents late in the day) will queue and complete under that policy. Standard support hours are 09:00–17:30 UK time, Monday to Friday (excluding English bank holidays), unless a separate support pack is agreed in writing.

Intellectual property & governing law

As between the parties, the Customer retains all intellectual property rights in Customer materials and in text drafts generated for the Customer through the Service (subject to the Provider’s underlying software, models, and tooling remaining the Provider’s IP). This Agreement is governed by the laws of England and Wales. The courts of England and Wales have exclusive jurisdiction over disputes arising out of or in connection with it.

Commercial add-ons (same schedule)

Seats mean named unique users on a published band: Boutique Team £449 / month flat (up to 5), Regional Practice £72 / named seat / month for 6–24 seats (monthly floor £521), Enterprise Operations £72 / named seat / month for 25–50 seats, or the then-current published rates. Each band includes a one-off onboarding & private node provisioning fee (currently £299 / £499 / £799 respectively). Concurrent inference uses the fair queue (FIFO lanes + round-robin) described in the performance schedule; latency is not guaranteed to match consumer chat products.
Outbound SMS auto-replies, where enabled, are template acknowledgements only. Contextual AI SMS is not included and will not auto-send.
The onboarding & private node provisioning fee covers organisation node setup on the Staffordshire host. Separately scoped Implementation & Security Audit work (office IP allowlisting, initial template parsing, Sovereign Data walkthrough) may be quoted in addition when agreed in writing.
Optional extras (extra seats beyond your band, persist-to-disk drafts, on-site walkthrough day) are quoted separately. Dedicated GPU or priority scheduling is roadmap-only until provisioned and is not included in the base organisation monthly fee.