First legal app certified “Works with Microsoft 365 Copilot

First in Copilot: shipping a legal-grade app for M365 Copilot

Microsoft’s Copilot certification is a gauntlet. Avishkaram engineered LawToolBox through it end-to-end — designing a matter-grounded agent architecture, navigating a 74-day evidence review, and shipping the first legal application to earn the “Works with Copilot” badge for Microsoft Teams.

Client

LawToolBox

Our Role

End-to-end engineering & certification partner

Platform

Microsoft 365 Copilot · Teams · Outlook

Certified

11 December 2023

1st

Legal app certified for M365 Copilot

74 days

Certification evidence review, end to end

4-stage

Copilot agent architecture we designed

0

Incremental engineering per new Copilot surface

— The Challenge

Copilot doesn’t understand privilege. We had to teach it.

When Microsoft opened M365 Copilot, it was powerful and general — and completely unsuited to legal work out of the box. Ask it for “upcoming deadlines” in a law firm and it would happily return a blend of court dates, internal meetings, and someone’s dentist appointment.

 

Worse, a flat tenant-wide index means Copilot can surface one client’s matter data in another client’s answer. For a legal product, that isn’t a UX problem — it’s a privilege problem.

 

LawToolBox needed a Copilot experience that was genuinely useful to attorneys and defensible under legal ethics rules. That meant grounding every answer in client-matter context, inheriting the firm’s existing compliance controls, and proving all of it to Microsoft.

The brief

Build a Copilot agent that answers legal questions with matter-level precision, keeps client data inside the customer’s tenant, and passes Microsoft’s “Works with Copilot” certification — ahead of every other legal vendor.

— What Avishkaram Did

We were the engineering and certification partner

This was a white-label delivery. LawToolBox owned the product and the brand; Avishkaram owned the architecture, the build, and the path through Microsoft’s certification process. Four workstreams, run in parallel.

Agent architecture & build

Designed and implemented a four-stage Copilot agent that resolves every query against matter containers instead of a tenant-wide index.

Certification leadership

Owned the Microsoft certification programme end to end — scoping controls before writing integration code, then driving the 14-day submission and 60-day evidence review to closure.

Privilege-first design

Every packaging, permission and grounding decision was made against legal privilege requirements, so client-matter data can never leak across matters.

Ecosystem leverage

Built natively on Microsoft’s agent framework so each new Copilot surface inherits the legal capability automatically — zero incremental engineering.

— The Architecture

Four stages from raw matter data to grounded answers

The core insight: don’t try to make Copilot smarter about legal data. Make the data arrive at Copilot already scoped to a single matter.

Matter containers

Each legal matter is provisioned as a dedicated Microsoft 365 group, with its own SharePoint site, Teams channel and calendar — an isolated data container per matter.

LawToolBox agent

A Copilot agent reads matter data — deadlines, documents, saved AI insights — and supplies it as grounded context for every query.

Grounded answers

Queries resolve against the matter container, so “upcoming deadlines” returns a clean legal list rather than a blended personal calendar.

Surface integration

The same agent works across Teams, Outlook and Copilot surfaces, enabling conversational queries and deadline reports straight from chat.

Containers, not indexes

The conventional approach is to crawl everything into one search index and filter at query time. We rejected that. A flat index makes cross-matter bleed a permissions question — and permissions bugs in legal software are catastrophic.

 

Per-matter containers make isolation structural. There is no shared pool of content to accidentally retrieve from.

Inherited compliance, not rebuilt

Client matters inherit Microsoft Purview sensitivity labels automatically. Rather than build a parallel security model that we’d then have to certify, we rode Microsoft’s own compliance infrastructure.

 

Matter containers stay inside the customer’s Microsoft 365 tenant. No data leaves for model training.

— The Gauntlet

74 days inside Microsoft’s certification process

Microsoft’s “Works with Copilot” review is not a formality. Our job was to make it predictable: scope every control before building, assign an owner and an evidence source to each one, and treat the review as an engineering workstream rather than an afterthought.

Phase 01 — Pre-build

Scope first, build second

We mapped every certification control before writing integration code, defining scope, owners, evidence sources and deadlines up front.

Phase 02 — 14 days

Initial submission

First formal submission to Microsoft, covering agent behaviour, grounding model, permission boundaries and data-handling posture.

Phase 03 — 60 days

Full evidence review

Sixty days of detailed evidence review across security, compliance, privilege handling and surface integration — run as a tracked engineering workstream with named owners.

11 December 2023

Certified — a first for legal

LawToolBox became the first legal application approved as “Works with Copilot” for Microsoft Teams, at a time when Copilot for M365 was available only to enterprises buying 300+ licences.

— Key Decisions

Four engineering calls that got us through

Each of these traded a little convenience for privilege safety — and each one made the certification review materially easier.

 
DecisionWhat Avishkaram did
Scope first, build secondMapped every certification control — owner, evidence source, deadline — before a line of integration code was written.
Containers, not indexesUsed per-matter Microsoft 365 containers instead of a flat tenant-wide search index, making isolation structural rather than permission-dependent.
Purview labels inheritedClient matters automatically inherit Microsoft Purview sensitivity labels, reusing Microsoft’s compliance stack instead of building a parallel one.
Ecosystem leverageBuilt natively on Microsoft’s agent framework so every Copilot surface expansion carries the legal capability automatically — zero incremental engineering per surface.

“We didn’t build a legal chatbot. We built the data boundaries that make a general-purpose AI safe to use in legal practice — and then proved it to Microsoft.”

Avishkaram Engineering

Copilot integration & certification team

— Beyond the Agent

What else shipped

Community prompt library

A shared library of legal prompts with firm-level customisation, so firms start from proven patterns rather than a blank chat box.

Automated matter provisioning

AI-ready matter containers are provisioned automatically from billing systems, iManage, NetDocuments and other legal platforms — no manual setup per matter.

Deadline reports in chat

Conversational queries and generated deadline reports available directly from Teams chat, without leaving the matter context.

Tenant-resident data

Matter containers remain inside the customer’s Microsoft 365 tenant. No client data is used for model training.

— The Architecture

Four stages from raw matter data to grounded answers

The core insight: don’t try to make Copilot smarter about legal data. Make the data arrive at Copilot already scoped to a single matter.

Category first

No legal vendor had cleared Microsoft’s Copilot certification before. Avishkaram defined the path and walked it first.

Privilege preserved

Matter-level grounding means Copilot answers are scoped to a single client matter — with Purview labels inherited, not reinvented.

Built to expand

Because the agent sits on Microsoft’s native framework, new Copilot surfaces inherit the legal capability with no extra engineering.

Reusable playbook

The certification approach — scope, owner, evidence, deadline — became a repeatable method for regulated-industry Copilot builds.