An AI workspace for support and engineering teams. It reads the ticket, searches the code, past tickets, and release notes, and hands support a brief they can act on. A person approves anything that touches a customer or a database.
Invoices stuck in pending after a mid-cycle plan change
When a customer upgrades mid-cycle, the proration step skips the final status update. The invoice is correct but never leaves pending.
- 01Tell the customer the amount is right and nothing was double-charged.
- 02Re-run finalization for the three affected invoices.
- 03Escalate the root cause to billing engineering.
- [1] billing/proration.py:212
- [2] billing/invoice_state.py:88
- [3] release-notes/4.12.md
UPDATE invoices SET status = 'finalized' WHERE id IN (8812, 8815, 8820) AND status = 'pending';
- 01
A fast model reads the ticket and sorts it: bug, how-to, or admin request. How-tos get an answer. Bugs get an investigation.
- 02
An agent searches similar past tickets, the codebase, docs, and release notes, inside a fixed tool budget so it always finishes.
- 03
Support gets plain English, what to do now, workarounds, and evidence with file and line citations. Engineering gets root cause and a fix plan.
- 04
Customer replies and database changes arrive as drafts. A person reads, edits, and sends. The Hub never writes on its own.
- 00:02Classified as bughigh confidence
- 00:09Pulled the ticketand 3 similar past tickets
- 00:21Searched the codeproration.py, invoice_state.py
- 00:34Checked release notes4.12 changed proration
- 00:41Drafted the brief3 sources cited
Paste a ticket key. It classifies, investigates, and drafts a support-ready brief with a copy-ready escalation note when the ticket needs engineering.
For tickets that reach engineering: likely root cause, where to look, recommended fix, validation plan, and regression risk. Run it quick, exhaustive, or let it decide.
One chat that can look up tickets, search code and past tickets, read release notes, run code, and search the web. Every claim cites its source.
The day's tickets, already sorted. Each one gets a one-line summary and a next action: investigate, waiting on customer, or close.
Queries written against the real schema and validated before anyone runs them. Reads stay read-only. Writes come with preview, verification, and rollback.
SLA breaches, stuck tickets, and backlog depth for leads. Who uses the Hub, how often they come back, and which tools earn their keep.
A finding in today's code is not called the cause until it ties to what is actually deployed. Low confidence says so out loud.
Database changes ship as drafts with rollback. Nothing reaches a customer without a person.
Every investigation has a tool budget and a defined stop. No runaway loops, no surprise bills.
Each step uses the model that fits it. Today that is Claude Sonnet for investigations and Claude Code for deep code reviews.
- Jira
- Microsoft Entra SSO
- Git repositories
- SQL databases
- Release notes
- Web search
- Models
- Claude, on your own Anthropic or AWS Bedrock account. Small, fast models handle classification and summaries.
We connect AZAI Hub to your ticket system, code, and data, then start with a 30-day pilot on one queue.
Book a demo