A multi-agent incident management platform shipped into a live federal enterprise environment supporting 30 plus U.S. Government customers. Nine agents and a Python automation pipeline working as one system, deployed inside M365 Copilot Agent Builder with full guardrails, SLA logic, and audit controls.
Support Escalation Managers were running 30 plus government accounts on manual review cycles. Triage, escalation routing, executive reporting, and risk prioritization were each their own slow, inconsistent process. Reviews that should have happened weekly often did not, and when they did, the answer depended on which manager was on shift that day.
IMHub started as a two-stage Python pipeline that took a four to six hour weekly case review down to fifteen minutes. From there it grew into a full agentic platform with executive PDF reporting, AI summaries, trend tracking, parent-child case relationships, audit logging, and an integrated feedback loop with IcM tracking.
Around the pipeline sits an ecosystem of nine M365 Copilot agents. Each one solves a real operational problem for an Incident Manager. They share a single entry point, IM OneStopShop, so users never have to think about which tool to reach for.
These are not demos. Every agent below ships with a full instruction set, edge case handling, SLA and cadence logic, RBAC-aware behavior, and human-in-the-loop governance.
Meta-agent that consolidates five distinct IM workflows behind a single interface. Silently classifies the request, applies the authoritative instruction set, and returns production-ready output. No routing logic exposed to the user.
Email triage agent. Reviews unread support and non-support email, groups by Tracking ID, and surfaces only threads requiring IM intervention. Enforces Sev A 4hr and Sev B 24hr engineering response windows so escalations stay out of engineering’s lane unless real signal is present.
Combined case review table from Outlook, Teams case chats, and uploaded files. Cadence evaluation built in. Sev B requires daily business-day updates, Sev C every three. Source evidence stays separated from the case table by design. Never guesses or fabricates.
Meta-agent that interviews an Incident Manager and builds their own custom Case Review Assistant inside Copilot Agent Builder. Ten phase structured process, ending with a real test conversation iteration before sign-off. Closes with custom icon design and upload guidance.
Monitors ServiceNow, IcM, and Iridis for active outages. Detects SLA risk and missing-update breaches. Always cites ticket IDs, timestamps, and owners. Never fabricates state or ownership. RBAC-safe by design.
Customer-facing Word document generator. Takes Reactive Support Summary screenshots or structured inputs and produces a formatted .docx with embedded screenshot, dominant incident drivers, and up to three Value-Based Delivery recommendations. Co-primary rule for products within 10% of the top case count.
Long-running case tracker. Registers cases with age tracking, flags inactivity, drafts escalation notices, and recommends specific next actions to progress stalled cases. Treats case, SR, and ticket as interchangeable.
Natural-language time logging for Teams. Short commands, an in-session ledger, end-of-day summary grouped by customer and entry type with rounding to the nearest 15 minutes. Session-scoped, so each day starts fresh.
The parts that keep the system honest in a production environment.
Enterprise tooling does not have to feel like enterprise tooling. The case review pipeline ships with a personality. Loading screens rotate practical IM tips and the occasional non sequitur. After 1,000 cases processed it unlocks a game of Snake and a personalized breakdown of the time the user has saved. At the end of every run, a Tip Your Developer Jar prompts users to select a donation amount, then immediately responds, "Just kidding, this tool is FREE."
People use tools they enjoy. Tools they enjoy get adopted. Adoption is what creates the 299 hours of annual savings, not the code alone.
Sanitized screenshots are added as the public set is approved. The source layout and instruction sets are visible in the linked repository.
Case review pipeline run, summary view.
IM OneStopShop classifying a multi-workflow request.
Executive PDF report with AI-generated summary and trend chart.
I am happy to demo IMHub or any of the agents in the fleet, share architecture detail, or talk through what it would look like to bring something similar into your environment.