Configuring the AI assistant for your workspace

Applies to: Workspace Admins (cloud and on-premise) · Time: 5 minutes for the built-in assistant, about 15 minutes to connect your company's own AI

Overview

The AI assistant answers questions about your workspace, summarizes records and translates text, in English, Russian and Uzbek. Your team uses it in three places:

• Ask AI: the button in the top bar (or Ctrl+J) opens a panel where people ask questions such as "Which opportunities close this month?". Answers link to the reports and records they came from.
• Summarize: on opportunities, accounts, tickets and projects, a short summary of where the record stands, the risks and the next step.
• Translate: on tickets and Knowledge articles, a translation into the reader's language.

The assistant only reads what the person asking can already see, through the app's own permission checks. It never changes your data.

You decide two things: where the answers come from (the AI provider) and who may use which features. Both are in Workspace settings → AI assistant.

Before you start

• You are a Workspace Admin.
• To use your company's own AI, have its details ready: the endpoint URL, the model name, and an API key if the service needs one. Your IT team or AI vendor can provide them.

Part 1: Choose the AI provider

Open Workspace settings → AI assistant. In the AI provider card, choose one of four options:

Optionally, enter an Assistant name, for example your company assistant's own name. It replaces "Ask AI" on the top-bar button and the panel, and labels summaries and translations.

Then click Test connection. The test sends one short request with the settings on the screen (nothing is saved or counted) and shows:

• Connected in … ms and a sample reply, when everything works;
• whether the model can read workspace data through tools. Models that can't still answer, but only general questions; they are told not to invent figures.

When the test passes, click Save provider. The assistant uses the new provider straight away.

Examples for OpenAI-compatible endpoints

Enter the address that ends in /v1; Digital Workspace adds /chat/completions. For the assistant to read workspace data, the model must support tool (function) calling.

On the cloud version the address must be public and use https://. Internal addresses are refused so nobody can point the assistant at internal systems. On-premise installations can use internal and http:// addresses.

Part 2: Connect your company's own agent

Choose Your own agent when your company already runs its own AI assistant. Digital Workspace sends each question, summary and translation request to the Agent URL and shows the agent's answer. Give your developers this summary; the full contract is also under What your agent receives and returns in the card.

1. Request. A POST with JSON: task (ask, summary, translation or test), instructions, messages (the conversation), workspace (name, slug) and user (name, language).
2. Answer. Reply 200 with { "answer": "…" }. Plain text also works.
3. Authentication. The API key is sent as Authorization: Bearer <key>, or in the header you set under Key header.
4. Checking the sender. When a key is set, every request carries X-DW-Timestamp and X-DW-Signature. The signature is sha256= plus the hex HMAC-SHA256 of timestamp + "." + request body, using the key. Your agent should check it and reject timestamps older than five minutes.

Let the agent read workspace data (off by default): when on, the request also lists the tools the built-in assistant uses (reports, records and help articles). Instead of an answer, your agent may reply with { "tool_calls": [ … ] }. Digital Workspace runs those tools as the person asking, with their permissions, and sends the results back as tool messages, until the agent answers. The agent never connects to your database. When the switch is off, the agent gets only the conversation, the text being summarized or translated, and who is asking.

Part 3: Decide who uses what

1. AI assistant on for this workspace: the main switch. When it's off, nobody sees the assistant.
2. Features: turn Ask the workspace, Summaries and Translation on or off separately.
3. Who can use it: Everyone in the workspace, or Only these roles. Workspace Admins can always use it.
4. Monthly requests: keep the plan's default, or choose Custom to set your own limit. The bar shows this month's use.
5. Click Save.

These rules apply whichever provider you chose.

Privacy and security

• The assistant sees only what the person asking can see. Salary is never sent. HR records are visible only to people with HR access, and confidential documents only to people who can open them.
• API keys are stored encrypted, shown afterwards only as ••••1234, and never sent to anyone's browser. To replace a key, type the new one; to remove it, tick Remove the saved key.
• The Audit log at the bottom of the page lists every request: when, who, which feature, which records were read, and which provider answered.

For IT teams: one setting for the whole server (on-premise)

On an on-premise server, IT can make the company's AI the Built-in option for every workspace with environment variables, then restart the app:

Workspace Admins can still choose a different provider for their own workspace.

Troubleshooting