Smart Biz is a Platform for Tailor-made software
AI workflows

AI that works on tickets and business data

Smart Biz connects to Google Gemini or OpenAI. The two built-in agents are Platform Chat AI and the Support Ticket Agent. Both keep a record of what they did, including token use.

Platform Chat AI

People ask questions from inside the application. The assistant is pointed at named datasets, each with a skill file: a description and keywords that tell the model when that data is relevant.

  • 1
    Ask from the screen

    Chat is available in the product. A page can offer sample questions so users start from a real task, not a blank box.

  • 2
    Match a dataset

    The question is matched to a dataset skill. The model then queries that dataset instead of answering from general knowledge alone.

  • 3
    Track the tokens

    Each call stores the provider, model, prompt tokens, and completion tokens against the user, so usage can be reviewed.

Smart Biz application screen
Support Ticket Agent

From an open ticket to a draft a person can send

The agent runs as a background job after someone requests it on a ticket. It does not email the customer on its own. The reply stays pending until staff approve it.

  1. Request

    A ticket in New or In Progress is marked for the agent. Closed and cancelled tickets are skipped. If AI is turned off, the request is cleared and nothing is drafted.

  2. Triage

    The model reads the ticket title, status, priority, page, URL, and the conversation. It returns an internal summary, a suggested type (Bug, Request, Question, or Other), a suggested priority, a confidence level, and three to five code-search queries.

  3. Search Bitbucket

    When the repository is configured, those queries are run against the source. The agent looks for the screen, document number, error text, or field named in the ticket. The search result is saved with the draft.

  4. Write the draft

    A second model call writes the customer reply in plain language and internal notes for staff: what was checked, the likely cause, and whether it should be escalated. Raw source code is not pasted into the customer reply.

  5. Approve

    The draft status is Pending. Support staff review it, then approve or retry. Only after approval does the reply go out as the ticket response.

  6. Close quiet tickets

    If a New or In Progress ticket has had no activity for the configured number of days, the agent can draft a close message and a short resolution note, then close the ticket and notify the reporter.

What is recorded

Every ticket-agent run opens an audit session linked to the ticket and the draft.

Steps

The session stores the triage prompt, the model reply, the Bitbucket search call, and the draft prompt.

Tokens

Prompt and completion tokens from both model calls are added up and saved when the session completes.

Failures

If a run fails, the draft is marked Failed, the error is stored, and the ticket can be retried.