Transforming Shared Mailboxes into AI-Powered Service Hubs

Bridge365 classifies and routes shared-mailbox email automatically - combining Microsoft Copilot Studio, an Azure-hosted backend, and rule- and ML-based classification into one auditable pipeline.

v0.6.2 BSUL-1.0 - Source-available Phases 1-3 Complete
15Connector actions
3Phases shipped
4Phases planned
1Onboarding wizard

What works today

Phases 1-3 are complete and in production use.

🔌

Custom Connector

Power Platform / Copilot Studio integration with 15 actions, including an autonomous-agent polling trigger for new messages.

Available
☁️

Azure Backend Service

Python Flask app on Azure App Service with Microsoft Graph API integration for reading, classifying, and drafting shared-mailbox messages.

Available
🧠

Classification Engine

Dataverse-backed rule table drives message routing today, with a defined path to AI-powered classification.

Available
✉️

Draft Creation and Routing

Reply drafting, category tagging, folder moves, delta queries, attachments, and extended properties - full Phase 3 routing pipeline.

Available
📜

Audit Trail

Every classification and routing decision is logged to Dataverse for compliance and traceability.

Available
🔒

Defense-in-Depth Security

Entra ID + Easy Auth OAuth2, email allowlist authorization, Key Vault secrets, and optional network restriction - three independent layers.

Available
🧙

Guided Onboarding

onboarding.ps1 auto-discovers your Azure/Power Platform context and walks through app registration, deployment, hardening, and Dataverse setup from one menu.

Available
📁

Mail Folder Discovery

GetMailFolders lets flows resolve destination folder IDs for MoveMessage, including nested department subfolders.

Available

What's planned

The roadmap for Phases 4-7 and general availability.

🔁

Phase 4 - Override and Approvals

Manual classification override, a state-machine workflow, and configurable approval matrices.

Planned - v0.7.0
🧩

Phase 5 - MCP Server Integration

A Model Context Protocol server exposing Bridge365 tools to GitHub Copilot CLI, with schema-validated tool documentation.

Planned - v0.8.0
🤖

Phase 6 - BART Classifier

Fine-tuned BART model deployed on Azure Machine Learning, with confidence-based fallback to rule-based routing.

Planned - v0.9.0
🛡️

Phase 7 - Production Hardening

Monitoring and observability, compliance logging, deeper security hardening, and disaster recovery / backup.

Planned - v0.9.x
🚀

v1.0 - General Availability

Production-ready release with full documentation, training material, and a defined support and maintenance model.

Planned - v1.0.0

How it fits together

Copilot Studio triggers the connector, which talks to an Azure backend that bridges Microsoft Graph and Dataverse.

Copilot Studio Cloud

Custom Connector
Autonomous Agent Trigger
NewMessageReceived (polling)

Azure Backend

Service Endpoint
Python Flask App
Microsoft Graph
Shared Mailbox
Dataverse
Classifications and Audit

Authentication is handled entirely by Microsoft Entra ID and Azure App Service's built-in Easy Auth (OAuth2 delegated flow) - no custom token-validation code. Authorization is a lightweight email allowlist check, with an optional network-level restriction as a third layer.

Explore the project