Your firm's knowledge,
made operational.

A private AI platform grounded in your documents, your systems, and your people. Hosted in Azure. Delivered under your brand.

Book a demo See how it works ›

Subscription assistants were never built for regulated work.

The moment a prompt leaves for a vendor's cloud, your firm loses custody of the data, control of the answer, and the audit trail behind it. Every user writes their own instructions. Nothing is centrally managed, measured, or branded.

01Chain of custody
Data from your firm is added to an individual subscription. That subscription opens on any mobile device and any laptop with a browser, stores the data in a personal account, and shares it with other people. Traceability ends as soon as the data is attached.
02Training exposure
A checkbox is your only assurance, backed only by the vendor's terms of service. AI providers have already had public incidents where data was retained or tracked when it was not supposed to be.
03Management
Seats, prompts, and integrations sprawl across vendors and users. No single place to control or revoke.
04Telemetry
Is the guidance aligned to your standards? Did a simple question run on the most expensive model and burn your tokens? Is every department adopting it, or are users silently not? Is AI output being passed to clients as finished work? Subscriptions cannot tell you.
05Standards and brand
Every user invents a methodology and a format. The firm's own templates never reach the answer.
06Hallucination and unguided search
Each assistant searches wherever it likes. One trusts a vendor site, another a public forum. Whatever it cannot find, it may invent, and you have no control over that answer.
PROBLEM 1 OF 6
Chain of custody
A client document enters the firm, then copies itself to wherever the personal subscription is signed in.
Your firm
Personal subscription
Personal
phone
Family
laptop
AI chat
Any browser
tab
✕ Traceability ends here
PROBLEM 2 OF 6
Training exposure
A checkbox and a vendor's terms of service are your only assurance.
VENDOR TERMS OF SERVICE
✓Do not use my data for training
"Client A's Q3 revenue is $4.2M…"
Public
model
"…a similar client reported $4.2M…"
Public incidents: data retained or exposed
PROBLEM 3 OF 6
Management
Five vendors, fifteen accounts, fifteen prompts. No single place to control or revoke.
ChatGPT
Claude
Gemini
Perplexity
Grok
ChatGPT
Claude
Gemini
Perplexity
Grok
ChatGPT
Claude
Gemini
Perplexity
Grok
Revoke all access✕ Not possible · 15 separate accounts
PROBLEM 4 OF 6
Telemetry
Subscriptions cannot answer any of these questions for you.
Aligned to your standards??
› Answer drafted from: public web
› Firm methodology applied: none
› Checked against standards: —
Right model for the task??
› Q: "When is a 1040 due?"
› Routed to: premium reasoning model
› 18,402 tokens for a one-line answer
Adoption across departments??
Tax
Audit
Wealth
IT
Ops
› usage by person: not reported
Reviewed before it reaches clients??
› Memo written 100% by AI
› Sent to the client 4 min later
› Human review: not recorded
PROBLEM 5 OF 6
Standards and brand
Three people ask for the same letter and get three different documents. None is on your template.
"Draft the engagement letter for Client B"
WEngagement Letter.docx
FileHomeInsertLayout
Prepared with AI Assistant
September 23, 2026
Client B
100 Main Street, Suite 400
ENGAGEMENT LETTER
We are pleased to confirm the terms of our engagement to prepare your 2026 federal and state income tax returns. This letter sets out the scope of services, our responsibilities, and yours.
ServiceFee
Federal and state returns$3,800
Sincerely,
Engagement Partner
Vendor template
Generic layout · no firm logo
WEngagement Letter.docx
FileHomeInsertLayout
H&A
Hartwell & Associates · 2023
September 23, 2026
Client B
100 Main Street, Suite 400
Letter of Engagement
We are pleased to confirm the terms of our engagement to prepare your 2026 federal and state income tax returns. This letter sets out the scope of services, our responsibilities, and yours.
ServiceFee
Federal and state returns$4,500
Sincerely,
Engagement Partner
Old letterhead
2023 fee schedule · retired terms
WEngagement Letter.docx
FileHomeInsertLayout
September 23, 2026
Client B
100 Main Street, Suite 400
engagement letter draft
We are pleased to confirm the terms of our engagement to prepare your 2026 federal and state income tax returns. This letter sets out the scope of services, our responsibilities, and yours.
fees: ?? (tbd)
Sincerely,
[name]
No brand at all
Plain text export · no template
PROBLEM 6 OF 6
Hallucination and unguided search
Same question, three answers: one guess, one dead end, and one citing a handbook section that does not exist.
"What is our time off policy?"
ASSISTANT 1
Employees accrue 15 days of PTO per year, prorated in the first year. Unused days roll over up to 5.
GUESS
ASSISTANT 2
I couldn't find that information. You may want to check with HR.
NO ANSWER
ASSISTANT 3
Per Handbook §9.4 (2024), PTO is unlimited with manager approval.
INVENTED
One governed platform solves all six✓ Your tenant✓ Never trains models✓ One admin console✓ Full telemetry✓ Your templates✓ Cited sources only

Every request, top to bottom.

A message leaves a client, crosses the edge, reaches the gateway, and is routed to the right specialist with the proper tooling and capabilities, which calls the model hub and searches the corpus while the intelligence layer feeds context in from the side.

01ClientsWeb app · Teams bots · iOS app · Office add-ins
02PerimeterPremium edge web application firewall in Prevention mode; every request crosses this edge
03GatewayRouter and relay · Routines · platform APIs
04OrchestratorThe SAXTech Orchestrator routes each turn to the right specialist, or a panel of specialists, with the proper tooling and capabilities. Any turn can reach SAXTech Resolve on the user’s own machine.
05Model hubAnthropic and OpenAI models hosted in Azure, behind a proxy with prompt caching
06Knowledge corpusAzure AI Search indexes with hybrid retrieval · session and router cache
07IntelligencePer-user Second Brain and the firm Master Brain enrich every turn from the side
WebTeamsiOSOffice
"What are the company holidays?"
Perimeter
Edge firewall · Prevention mode
Gateway
Router · relay · Routines
SSO ✓Rate limit ✓Route
Orchestrator
Routes to the right specialist with the proper tooling and capabilities
Model hub
Anthropic · OpenAI · hosted in Azure
ReasoningFast
Knowledge corpus
Hybrid search · citations
Handbook §6.2HR policy
Intelligence
Second Brain · Master Brain feed context
Boston officePrefers conciseFinance
✓ Answer streamed back with citations

Seven layers. One request path.

A request must clear every applicable layer before it reaches data, and a failure at any one layer stops it. Consequential boundaries are enforced by infrastructure the model cannot argue its way past.

✓Hosted in SAX-owned Azure, US regions
✓Your data never trains foundation models
✓GLBA-aligned handling of client financial data
✓IRC §7216 controls built in
✓Full per-user erase on request
✓Every answer cites its source
Read the security summary ›
01
Perimeter
Edge web application firewall in Prevention mode on every hostname.
Verified
02
Identity
Entra SSO with MFA. No local passwords. Re-checked on every turn.
Verified
03
Transport
TLS everywhere. Managed identity to every backend. No shared API keys.
Verified
04
Authorization
Group-gated specialists. Admin surfaces gated at the edge and the application.
Verified
05
Isolation
Per-user data stays with its owner. Shared knowledge crosses only after de-identification.
Verified
06
Data tiers
Regulated material is tier-scoped by identity, down to the field.
Verified
07
Governance
Human publish gates, refuse-by-default classifiers, confirmations, and an audit trail.
Verified
Untrusted traffic BLOCKED
Request cleared all seven layers · data released

Everywhere your people already work.

One platform, one identity, one set of specialists, reached from four surfaces. Voice is available on iOS and the web.

Web app
Full experience: specialists, projects, shared conversations, live document previews, Second Brain.
Microsoft Teams
A bot in channels and 1:1 chats that switches specialists automatically and collaborates with the people in the thread.
9:41▂▄▆ ▭
Tax Specialist
Tax & Accounting
▾
✦ Show Me How🎓 Learn⊕ IT Support▦ Portals⚠ Report
Is the R&D credit refundable for us this year?
REASONING
✓Searching firm memos · 4 sources
✓Checking IRC §41(h) thresholds
✓Reading your 2025 return
Yes, partially. You qualify as a small business, so:
•Under $5M gross receipts, first 5 years
•Up to $500k offsets payroll tax each year
•Remainder carries forward 20 years
1 Firm memo2 IRC §41(h)3 2025 return
Draft a client email explaining it.
Handed to Client Comms · context kept
W
R&D Credit · Client Email.docx
Firm letterhead · cites 3 sources
OpenSend via OutlookShare
+
Message Tax Specialist…〰
↑
iOS app
Native, signed in with your identity, streaming answers, documents, and voice with barge-in.
Office add-ins
MegaSnipper for Word, Excel, PowerPoint, and Outlook: OCR, create and edit in place with live previews, attach conversations to shared documents.

Industry agnostic by design. Specialized by configuration.

The specialists, prompts, skills, and corpus are built for your practice areas during implementation. The platform underneath does not change.

Tax
Wealth
Audit & Attestation
General Corporate
Finance
Mergers & Acquisitions
Client Accounting Services
Information Technology
Construction
Real Estate
Insurance
Nonprofit

Currently in production at SAX.

The platform began as the internal AI for SAX, a Top 100 accounting, tax, and advisory firm, where it serves tax, audit, wealth, client accounting, M&A, and technology teams across multiple entities. SAX Technology Advisors just productized the same methodology, architecture, infrastructure to bring to market an enterprise class SaaS AI platform to any business looking to consolidate, standardise, or implement AI vs utilizing various existing tools that are adding chatbots or users set up with individual/teams public subscriptions.

SAX’s own results · live from its ROI dashboard · 5 months in service
Questions answered
133,922
Hours returned
9,458
Modelled from classified work across every team.
Firm adoption
82.3%
Return on investment
14.3×
Total ROI over 5 months in service.
Documents created for staff
17,826
X
8,779
XLSX
W
5,186
DOCX
2,490
PDF
P
927
PPTX
Documents ingested and analyzed
51,308
X
9,657
XLSX
W
4,307
DOCX
27,929
PDF
P
441
PPTX
At SAX, every $1 invoiced has returned $14.26 in staff capacity

These figures describe SAX’s own deployment only and are not a projection or guarantee of results for any other organization. Cost is from Azure invoices; value is measured usage multiplied by SAX’s stated assumptions for time saved and loaded labor rates.

See it on your own data.

Every engagement starts with a demo and a consult. Pricing follows the scope: implementation, hosting and maintenance, per-user licensing, and usage true-ups.

Book a demo How engagements work ›