Comagine Health Product & Technology
Product & Technology — AI Hub

AI @ Comagine Health

Your hub for all things AI at Comagine Health — approved tools, use cases, and policy. From enterprise-wide baseline tools to advanced platforms for power users and technical staff, use this to understand what's available to you and how to request access to higher tiers.

May 2026 Internal Use Only
Read the AI FAQ → Approved Tools →
✓
Approved & availableThese tools are cleared for staff use on the approved use cases below.
@
Use your work accountITS-provisioned @comagine.org in enterprise mode — no personal or free accounts.
◉
Keep a human in the loopA qualified person reviews and approves every AI output.
+
New use cases require assessmentNew tools or high-risk use cases go through the AI Impact Assessment.
Restricted Data & HIPAA: Restricted or sensitive data — including PHI, PII, client-confidential, financial, and proprietary information — may only be entered into an AI tool when (1) that tool is approved on the AI Inventory for that data classification, (2) the use case appears in the Approved AI Use Cases catalog (or an AIA-approved department procedure), and (3) the use complies with applicable contracts and regulations. Absent these approvals, use only de-identified, non-sensitive, or public information. No tool listed here is HIPAA-compliant by default — a BAA and ITS-approved configuration are required before any PHI is processed.

Approved AI Use Cases

All tools in this guide are approved for these use cases under AI Policy ITS-0022 (Section 5.1.1):

All permitted use cases are conditional upon adherence to data security and governance requirements, including required human review and editing. This list is not exhaustive; new or high-risk use cases are added through the AI Impact Assessment (AIA) process.

Tools by Access Level

Three access levels, from the baseline available to everyone, up to advanced and agentic platforms for technical staff.

Baseline — All Staff
Available to all employees. Part of existing Microsoft Office 365 E3 licensing. Provides a capable AI chat experience with commercial data protection — suitable for general Q&A, drafting, and summarization. Note: this is a web-based chat interface only and does not include deep integration with Word, Excel, Teams, or Outlook (those features require the Advanced Copilot add-on).
Tool Model(s) Available What It's Good For Key Capabilities Limitations
Microsoft Copilot Chat Included in O365 E3
  • GPT-4o — Primary chat model
  • GPT-4o mini — Faster lightweight tasks

Commercial data protection enabled — prompts and responses are not used to train Microsoft or OpenAI models.
  • General Q&A and research assistance
  • Drafting emails, memos, and short documents via chat
  • Summarizing pasted text or content
  • Brainstorming and ideation
  • Basic coding help and formula suggestions
 Commercial Data Protection Web Interface GPT-4o Quality No Additional License Needed Chat Interface Only No Word/Excel/Teams Integration No Meeting Transcription No File Upload/Analysis
Advanced — Request Access
Available on request — approved by department lead and Product & Technology. Includes the full Microsoft 365 Copilot add-on (deep Office integration), Google Gemini through our Google Workspace for Nonprofits plan, and Claude Enterprise. (Google NotebookLM and other tools are on the Approved Tools list.) These tools significantly expand capability for staff with regular, structured AI workflow needs. These tools are approved and available to staff for the approved use cases, conditional on data security and governance requirements; ITS provisions accounts on request.
Tool Model(s) Available What It's Good For Key Capabilities Limitations
Microsoft 365 Copilot Full Add-On License
  • GPT-4o — Embedded across M365 apps
  • GPT-4o mini — Lightweight Office tasks
  • o1 / o3-mini — Reasoning (select surfaces)
  • Teams: live meeting summaries, action items, follow-up drafts
  • Word: draft and rewrite directly inside documents — SOWs, reports, proposals
  • Excel: formula generation, data insights, chart suggestions from natural language
  • Outlook: email drafting, thread summarization, scheduling assistance
  • SharePoint: search and surface information across internal document libraries
 Deep M365 Integration Teams Meeting Summaries Word / Excel / Outlook Embedded SharePoint Grounding Additional License Required Best Fit Within M365 Ecosystem PHI: M365 HIPAA BAA Required
Google Gemini Google Workspace for Nonprofits
  • Gemini 1.5 / 2.0 — Available in Workspace apps

Verify exact model tier included in your nonprofit plan — advanced models (2.5 Pro) may require an upgraded plan. Contact Product & Technology to confirm.
  • Drafting and editing in Google Docs with AI assistance
  • Slide generation and content suggestions in Google Slides
  • Smart replies, drafting, and summarization in Gmail
  • Formula help and data analysis in Google Sheets
  • Long-document summarization with extended context window
 Google Workspace Embedded Docs / Slides / Gmail / Sheets Extended Context Window Exact Model Tier TBC — Verify with Google PHI: Google Workspace BAA Required
Claude Enterprise Anthropic
  • Claude Sonnet 4.6 — Primary (balanced, fast)
  • Claude Haiku 4.5 — Lightweight / high-volume

Conversations are not used to train Anthropic models by default on Team plans.
  • High-quality consulting deliverables — proposals, gap analyses, status reports, SOWs
  • Uploading and querying long documents (200K token context window)
  • Nuanced client communications requiring careful tone and precision
  • Code review, scripting, and lightweight EHR/API coding tasks
  • Projects: persistent workspaces that retain context across an entire client engagement
 Best Writing Quality 200K Token Context Persistent Projects Strong Document Analysis Shared Workspaces No M365 / Google Workspace Integration No Image Generation PHI: Claude for Enterprise + BAA Required
Agentic — Power Users & Builders
For power users, technical architects, and senior staff with specialized needs. Includes agent-building platforms, advanced reasoning models, and enterprise-grade AI. Access requires Product & Technology approval and a documented use case. Higher complexity — not recommended for general staff without specific workflows that justify the upgrade. Agentic and automated workflows require an approved AIA, human-on-the-loop supervision, and written authorization from the responsible department VP and the VP of Product & Technology; autonomous public-impacting decisions are prohibited.
Tool Model(s) / Capabilities What It's Good For Healthcare IT Strengths Limitations
Microsoft Copilot Studio + Azure AI Foundry
Microsoft Models
  • Phi-3 / Phi-3.5 — Lightweight on-device SLMs
Via Azure AI Foundry
  • Meta Llama 3.x — Open-weight, flexible
  • Mistral Large / Small — EU data residency option
  • GPT-4o / o-series — Via Azure OpenAI Service
  • 1,600+ models — Full Azure AI catalog
  • Build custom HIT agents — FHIR terminology lookup bot, prior auth summarizer, referral triage assistant
  • Automated multi-step workflows across M365, SharePoint, and third-party systems
  • On-premises or data-residency-constrained deployments using open-weight models
  • Internal-facing chatbots grounded in SharePoint or Confluence knowledge bases
 Custom Agent Builder Open-Weight Model Support On-Prem / Data Residency Options M365 Workflow Automation

Most flexible platform for building production HIT tools. Data-sensitive projects can use Phi or Llama on Azure without data leaving the tenant — important for client-specific deployments.

 Requires Developer / Architect Skill Complex Pricing Model Agentic Use: AIA + VP P&T Sign-Off PHI: Azure HIPAA BAA + Compliance Review
Gemini Advanced
+ Deep Research Google One AI Premium
  • Gemini 2.5 Pro — Most capable, advanced reasoning
  • Gemini 2.0 Flash — Speed-optimized
  • Deep Research mode — Autonomous web research agent
  • NotebookLM Plus — Enhanced source capacity
  • Deep Research: autonomously searches the web, synthesizes multi-source findings into a structured report with citations
  • Complex reasoning over large regulatory corpora — HIPAA, HL7, FHIR specs in one pass
  • Advanced coding and API integration tasks
  • NotebookLM Plus: larger document capacity for extensive project libraries
 Deep Research Agent 2.5 Pro Reasoning 1M Token Context Grounded in Live Web

Deep Research is the standout feature here — best in class for competitive intelligence, compliance gap analysis, and regulatory landscape research with minimal manual effort.

 Upgrade from Nonprofit Plan Required PHI: Google Enterprise BAA Required
Claude Enterprise Anthropic
  • Claude Opus 4.6 — Most capable model
  • Claude Sonnet 4.6 — Balanced (also in Advanced)
  • Extended Thinking — Visible reasoning chain
  • Computer Use (API) — Desktop automation agent
  • High-stakes consulting deliverables requiring deep, structured analysis
  • Extended Thinking: model shows step-by-step reasoning before answering — valuable for auditable HIT recommendations
  • Complex coding — Epic/Cerner API integrations, FHIR resource logic, custom HL7 parsers
  • Computer Use: automate repetitive desktop workflows via agent (API-level feature)
  • Enterprise: SSO, admin controls, priority access, custom retention policies
 Auditable Reasoning (Extended Thinking) Advanced EHR / API Coding Computer Use Agent Enterprise Admin Controls

Extended Thinking is particularly valuable in HIT consulting — showing the reasoning chain helps justify complex recommendations to clinical or compliance stakeholders, and creates an auditable decision trail.

 Opus Slower Than Sonnet Computer Use (Agentic): AIA + VP P&T Sign-Off PHI: Claude for Enterprise + BAA Required
ChatGPT Enterprise OpenAI — o-series Reasoning
  • GPT-4o — General flagship
  • o1 — Deep deliberate reasoning
  • o3 — Most advanced reasoning model
  • o4-mini — Fast reasoning at scale
  • Deep Research — Autonomous research agent
  • Code Interpreter — In-session data analysis
  • Complex multi-step HIT architecture reasoning — EHR integration design, FHIR/HL7 mapping, interoperability planning
  • Deep Research: autonomous multi-source research with cited reports
  • Code Interpreter: analyze de-identified exported datasets directly in-session
  • ICD/CPT code analysis, prior auth logic evaluation, regulatory gap identification
 Strongest Reasoning Models (o3) Deep Research Agent In-Session Data Analysis Broadest Model Selection

The o-series reasoning models — particularly o3 — are best-in-class for complex, multi-step technical problems. Recommended only where sustained high-level reasoning on HIT architecture or compliance logic is a regular workflow need.

 Highest Cost — Justify Use Case First o-series Slower Than Standard Models PHI: ChatGPT Enterprise + BAA Required
How to Request Access

BBaseline

Already available. Log in at copilot.microsoft.com with your O365 credentials. No action needed.

AAdvanced Access

Submit a HelpDesk ticket (freshservice@comagine.org) with your intended use case. ITS provisions your account for the approved tools. New or high-risk use cases beyond the approved list go through the AIA process.

+Agentic Access

Requires a documented use case and an approved AIA. Agentic or automated workflows also need human-on-the-loop supervision and written sign-off from your department VP and the VP of Product & Technology.

✓Approved Use Cases

These tools are approved for the use cases listed at the top of this guide, conditional on data security and governance. Confirm your task fits an approved use case; new or high-risk use cases are added via the AI Impact Assessment (AIA).

§Client Contracts

For client deliverables, obtain and follow each client's AI policy — some prohibit AI use. Where a client policy conflicts with ours, contact Compliance via HelpDesk before proceeding.

!Restricted Data & HIPAA

PHI and other restricted data may only be used in a tool approved on the AI Inventory for that classification, under an approved use case, with a BAA in place. Otherwise use only de-identified or non-sensitive data.

?Not Sure Which Group?

Start with Baseline. If you find yourself hitting limits or simply need a consultation — contact us and we can help!

$Licensing

Licenses are procured by ITS and may be covered by an indirect rate or a specific contract — you don't need to budget for it personally. Submit a HelpDesk ticket to request access.

iAI FAQ

Common questions answered in plain language — what you can use, what's allowed, and key terms. Open the AI FAQ →

TApproved Tools

The full list of approved AI tools — including meeting assistants and tools with embedded AI. See Approved Tools →

Prepared by Comagine Health Product & Technology  |  Aligned with AI Policy ITS-0022  |  For Internal Distribution Only  |  May 2026
The AI Inventory is the authoritative source for approved AI tools, models, and use cases.
Model availability, features, and pricing subject to change. Verify current details with each vendor before finalizing procurement decisions.
Nonprofit plan inclusions should be confirmed directly with Microsoft and Google.