Workspace Assistant: AI Integration

The workspace assistant is the hub’s AI chat for how-tos, drafts, and scoped reports. TenantAdmins turn it on and wire the model under Admin → Integrations. You open it from HUB → Assistant, pick a module scope, and send — with API keys never in the message box.

What the workspace assistant is

The workspace assistant is a signed-in hub app: AI chat that sits beside CRM, projects, inventory, accounting, blog, and the other everyday tools. You use it for product how-tos, brainstorming, Quill-ready blog drafts, and table-first HTML reports — without leaving inveazy for a separate chatbot.

It is not a second CRM or warehouse screen. It does not run SQL or write sales orders for you. When configuration allows, the server can attach a live data snapshot for the module scope you picked so answers cite real workspace rows instead of guessing. Cross-module guesses can carry a [VERIFY] marker — check the source screen before you act on those lines.

On screen you get three parts: module scopes on the left, a plain-text message composer in the center (no file upload in chat), and prompts / model / rules on the right. A premade prompt is a templated starter on that right rail — swap placeholders (project name, account, warehouse) and send. Ask for a report, status summary, or one-pager when you want formatted HTML you can print or paste.

Why this shape: A chat that invents pipeline dollars or on-hand counts is worse than silence. Scopes, optional read-only data, locked report formats, and keys in Integrations are how the assistant stays honest.

How it is configured

Two layers matter. First, the app must be enabled for the organization and allowed on your role (same Applications + role pattern as other hub apps). Second, a TenantAdmin wires the model under Admin → Integrations:

1. Applications — turn on Workspace assistant
2. Providers — paste bring-your-own (BYOK) API keys; they are encrypted after save and never shown again
3. AI assistant — enable AI for the workspace, choose a primary provider and optional default model, optionally allow read-only workspace data, then save

Supported BYOK paths include OpenAI, Anthropic, Google Gemini, Azure OpenAI, Azure Foundry (Claude), and Ollama for local models. The host may also offer a built-in endpoint. You can still pick a model on the assistant’s Model tab; admins can save a workspace default. Organization context from Admin → My organization can inform replies. Workspace rules (tone, terminology, boundaries) apply to everyone in the tenant.

Never paste API keys, passwords, or card numbers into chat
Keys belong only in Admin → Integrations → Providers

This post is the product deep dive — what the assistant is, how you configure it, and how scopes and reports behave. For click-by-click menus, keep the sister how-to Blog, assistant & files open beside you. The wider Integrations map is in Accounting & integrations.

Operator path

HUB → Assistant → Workspace assistant
↓
Pick module scope (CRM, SCM, PM, …)
↓
Insert premade prompt or ask for a report
↓
Read / print / copy HTML — check [VERIFY] when you see it

Secrets: Admin → Integrations (Providers + AI assistant tabs)

So remember: always pick a real module scope when you need live numbers. That is what keeps the counts honest.

Example: You pick Inventory (SCM) scope, insert What to pick today?, and get real warehouse numbers — because read-only data is on. Ask the same question on General with read-only off, and you should get a refusal to invent counts, not a fake pick list. That is the difference.

Where it lives in the hub

When the app is enabled for your role, you can open it several ways:

1. HUB → Assistant → Workspace assistant — megamenu (fastest everyday path)
2. Hub overview — Workspace assistant shortcut tile
3. Blog hub → Review in assistant — opens with Blog scope ready (Blog, assistant & files)
4. Dashboards → Analyze / Analyze P&L — deep-link with a scope and starter prompt
5. Settings / blocked state — links into Admin → Integrations; wider map in Accounting & integrations

You land on a three-part workspace: scopes on the left, the message composer in the center, and prompts / model / rules on the right. The composer is plain text only — no image or file upload in chat. Attachments belong in File manager. Keep messages under the app length limit (about four thousand characters).

Module scopes (authoritative map)

Before you ask about live work, pick a module scope. You only see scopes for apps you can open and that your TenantAdmin enabled. Changing scope does not rewrite earlier replies — treat a switch as a fresh ask.

General is for product how-tos and brainstorming. It does not attach live database rows. For pipeline, warehouse, AR, or project status that should cite real work, choose the matching scope below and make sure read-only workspace data is on.

This is the one scope map to trust — purpose plus a sample premade prompt for each row:

Counter POS and Scheduling are not separate scopes — till-related questions ride Inventory and Accounting when those modules are in play. The model does not run SQL and does not write business rows; the server attaches a live data snapshot when policy allows.

Tip: Without read-only data and a real module scope, do not treat any dollar or quantity in the reply as operational truth.

Setting up BYOK keys under Integrations

The configuration steps are in How it is configured above. In short: Applications → Providers → AI assistant under Admin → Integrations, with optional read-only workspace data so scoped answers can cite live rows.

Never paste API keys, passwords, or card numbers into chat
Keys belong only in Admin → Integrations → Providers

Full Integrations map (Stripe, QuickBooks, email, and more): Accounting & integrations. After you sign in, Help covers Assistant overview and Workspace assistant.

Premade prompts and HTML reports

On the right rail, open Prompt templates… (and the Built-in area next to it). Templates filter by the scope you selected — use the sample prompts in the Module scopes table above, then swap placeholders and send.

When you ask for a report, status summary, executive summary, quick report, or one-pager, you can get formatted HTML instead of plain chat: title, date, numbered table of contents, KPI cards, status badges, tables first, short prose after each table, and numbered recommendations. Print it or copy the HTML into email. If a line is marked [VERIFY], check the source screen before you treat it as final.

Built-in locked formats include Report / export — rich HTML output and Blog draft — Quill-ready HTML. Workspace rules (tone, terminology, boundaries) sit beside those and apply to everyone in the tenant.

Pick scope → Prompt templates… → Insert → swap placeholders → send

Capabilities and boundaries

Capabilities

You can get inveazy how-tos in plain language. You can draft and brainstorm copy, including Quill-ready blog HTML in Blog scope. You can ground answers in live data snapshots when that option is on and you picked a real scope. You can produce printable, copyable HTML reports with KPI cards and recommendations. You can use organization context from Admin → My organization. You can clear your own chat history without changing Integrations settings.

Boundaries

The assistant cannot replace CRM, PM, or SCM screens, or export a full grid the way those apps do. It must not invent live counts on General or when read-only data is off. It does not execute SQL and does not write or update business documents. It does not accept files or images in the composer. It is not a vault for passwords, card numbers, or API keys — those belong only in Integrations. Those boundaries are the product contract, not optional etiquette.

Questions we get

Why did the assistant invent a stock number?

Usually you were on General, or read-only workspace data was off. Switch to Inventory (SCM), confirm the AI assistant tab allows read-only data, and ask again — or open the warehouse screen and treat the UI as source of truth.

Where do I put the OpenAI (or other) key?

Admin → Integrations → Providers, then save AI assistant settings. Never paste the key into the message box. See Accounting & integrations for the wider secrets map.

Can the assistant update a sales order for me?

No. It reads when policy allows; it does not write operational documents. Change orders in SCM (or CRM, PM, Accounting) yourself.

What is the difference between a chat reply and a report?

Ask for a report, status summary, executive summary, quick report, or one-pager when you want table-first HTML with KPI cards and recommendations. Everyday how-to questions stay as normal chat.

Bottom line

The hub workspace assistant earns trust when you pick the right scope, read-only data is on, and BYOK lives in Integrations. Premade prompts and report asks turn that trust into printable HTML. It is not a second database, not a key vault, and not a replacement for the hub screens that own the work.

The contract: Real scope → optional live snapshot → report when you need tables. Secrets stay in Integrations. Chat is not a second database and not a key vault.

Click-by-click menus: Blog, assistant & files. Platform architecture: technicals series. After you sign in, open Help → Assistant overview or Workspace assistant anytime.

Help → Assistant overview · Workspace assistant

Ready to try it? Explore a live demo at demo.inveazy.com, or see Features & pricing when you want your own site.

Glossary: Workspace assistant — hub chat under Assistant. Module scope — product area for the ask (see table above). General — help with no live DB rows. BYOK — bring-your-own key, encrypted under Integrations. Live data snapshot — read-only rows the server may attach for a non-General scope when allowed. Premade prompt — templated starter; swap placeholders and send. [VERIFY] — check the source screen before acting.