Aroma takes its platform into Slack with NetSet Software's conversational AI engineering.
NetSet Software built a conversational AI bot that turns natural-language messages into ticket actions and admin-approved access changes, with security designed for multi-tenant SaaS.
AROMA
USA · SaaS / Customer Support
Identity & access actions with confirmation and admin approval
At a glance · An AI & integration engagement
slash commands, @mentions, direct messages and channel listening
Create, update and close workflows through commands, mentions and direct messages.
Azure AD and Stripe user provisioning, with admin approval on each member request.
Rate limits, prompt-injection filtering and access controls for each organisation.
The partnership
A ticketing platform. A conversational layer built to match it.
Aroma is a multi-tenant ticketing and customer-support platform. Organisations use it to manage boards, tickets, assignees and workflow states, and many of their teams coordinate daily work in Slack.
Aroma asked NetSet Software to extend the platform into Slack. Users needed to act on tickets and request access changes from Slack, and Aroma's data model, permissions and tenant boundaries had to stay intact.
The result was more than a notification bot. It is a conversational AI interface connected to Aroma's core ticketing system and to two external identity platforms.
The challenges we tackled
Making AI useful without making it risky.
Taking actions inside Slack brings engineering problems that a web portal does not face.
01 / Unstructured input
Turning conversation into reliable data
Slack messages are informal, short and full of formatting such as links and mentions. The bot had to pull out structured fields (board, priority, state, assignee) and map them to real records, not free text.
02 / Multi-tenancy
One bot, many isolated organisations
Aroma keeps each organisation's data in its own database schema. Every Slack interaction had to be resolved to the right organisation, and users who belong to several workspaces had to be able to switch between them.
03 / High-impact actions
Access changes need oversight
Inviting, disabling or resetting users in Azure AD and Stripe has real security consequences. Members needed a way to request these changes without being able to run them unilaterally.
04 / AI-specific threats
Prompt injection and abuse
An AI interface open to every workspace member needed protection against prompt-injection attempts and overuse, and administrators needed to see and act on suspicious activity.
Our engineering approach
Confirm first. Approve where it matters. Scope everything.
NetSet Software designed the bot around three principles that apply to every way a user can reach it.
Human in the loop
AI proposes, people decide
Any action the AI interprets is shown back to the user as a preview with Confirm and Cancel buttons. Nothing is written until a person approves it.
Approval routing
Sensitive actions go to admins
Identity actions requested by members become approval requests. They are sent to every linked admin, claimed atomically so only one decision takes effect, and expire automatically after 24 hours.
Tenant scoping
Isolation at every layer
Workspace installations, user links, security settings, rate limits and conversation history are all stored per organisation, so no context crosses tenant boundaries.
The bot meets teams where the conversation already is, and asks before it acts.
NetSet Software × Aroma
What we built
A conversational interface connected to three systems.
Ticket workflows in Slack
Slash commands open forms that adapt to the selected board's priorities and assignees. They cover creating, updating and closing tickets, and enforce Aroma's rule that a ticket needs an assignee before it can be closed.
AI mentions and conversational DMs
Natural-language instructions in channels and direct messages are interpreted with OpenAI function calling. Direct messages keep a short conversation history for each workspace, so users can ask follow-up questions and look up tickets.
Ambient channel suggestions
In channels where it has been added, the bot can spot a message that reads like a ticket request and privately offer to act on it. Only the message's author sees the offer.
Identity provisioning
Fifteen Azure AD and Stripe actions are available from Slack: user invitations and offboarding, group and application access, account creation, and password and MFA resets.
Security and governance controls
Per-user rate limits, prompt-injection screening, organisation-defined keyword filters, user blocking with admin alerts, a security event log, and per-organisation feature toggles.
Notifications
When users are assigned a ticket, or a ticket's status changes or it is closed in Aroma, they get a direct message, triggered from the platform's own ticket events.
Engineering outcome
A governed, conversational surface for the platform.
Aroma available where teams already work
Ticket work and access requests can start where they come up in conversation, not later in a separate tool.
Automation with accountability
Every AI-proposed action needs explicit confirmation, and every member-initiated identity change needs admin approval, recorded with who decided and when.
Built for many organisations
The integration follows Aroma's multi-tenant model, so each organisation has its own Slack installation, settings, rules and AI key.
A pattern for further integrations
The parse, confirm, approve and execute pipeline gives Aroma a repeatable way to bring more platforms and actions into Slack.
The outcome is an AI interface that extends Aroma into Slack without giving up control over data, access or tenant boundaries.
Technology behind the Aroma Slack bot
A Python stack built for reliability and isolation.
Slack Bolt + Socket Mode
Event handling without public webhooks
The bot runs on Slack's Bolt framework over Socket Mode and reconnects automatically, with OAuth-based installation and token storage for each workspace.
OpenAI function calling
Structured intent extraction
The AI layer uses function calling to turn messages into typed actions. Results are checked against live board, state and member data before anything is proposed. Each organisation supplies its own API key.
Django + django-tenants
Schema-per-tenant data isolation
The integration is built into Aroma's Django backend. Every Slack request runs inside the correct tenant schema.
Microsoft Graph and Stripe SCIM 2.0
Standards-based identity provisioning
Azure AD operations use the Microsoft Graph API with client-credential authentication. Stripe Dashboard users are managed through Stripe's SCIM 2.0 API.
Redis and Celery
Rate limiting, context and scheduled expiry
Redis holds rate-limit windows and conversation history for DMs. Celery schedules approval requests to expire.
NetSet Software's contribution
End-to-end ownership of the integration.
NetSet Software designed, built and tested the Slack bot inside Aroma's existing platform:
Validation and production readiness
Tested at the points where things go wrong.
The integration ships with 135 automated tests across 15 test modules. They focus on the areas where errors would be most costly:
Approval flow
Request creation, admin routing, atomic claiming and the handling of requests that are already decided or have expired.
Identity operations
Azure AD user creation, group management, password resets and how each action is dispatched from Slack.
AI parsing
Intent extraction and how well it handles real-world Slack message formatting.
Linking and installation
Linking user accounts across tenants, storing workspace installations and the OAuth callback.
Commands and notifications
Slash command handling and ticket-event direct messages.
Other safeguards: the bot reconnects automatically if the connection drops, a global error handler logs failures and tells the user, and admin alerts are best-effort so they never block the main request.
AI & Integrations
Bring your platform
into the conversation.
Conversational AI with the confirmation, approval and isolation controls your product needs.