Back to Systemly

AI

The assistant

What it can see, what it can do, and how to ask well.

What counts as AI input

AI input includes your message, recent conversation history, workspace instructions, the assistant’s system rules and available tool descriptions. The system rules use compact wording while retaining permissions and action limits. Long instructions and large tool results can still increase input. Use New chat for a new topic and keep workspace instructions focused. Starting a new chat leaves the previous saved conversation available.

Start a conversation

  1. Open AI Studio in a client workspace, press Command + J (Control + J on Windows), or choose Help → Ask the assistant while working.
  2. Ask a specific question or describe the draft or change you need.
  3. Review its answer and the tool activity shown in the conversation.
  4. Use New chat for a fresh topic, or return to one of your saved conversations.

Conversations belong to you inside that workspace. A teammate cannot open your chat transcript, and switching client workspaces changes the data the assistant may use.

Keep or reset your quick chat

Closing the Command + J quick chat keeps the current conversation and draft for 15 minutes. Reopening it during that time resumes the same chat, including when you move between pages in the workspace. Help → Ask the assistant opens this same quick chat in Product help mode, when platform support AI is configured. It answers from the public documentation on every plan and uses no included AI allowance. Command + J or the Ask AI button returns to workspace AI. Select New chat at the top right to start fresh immediately. After the panel has been closed for 15 minutes, its next opening starts fresh. Reloading the browser also starts a fresh quick chat; saved conversations are available in AI Studio.

Speak to a human

Ask “I need to speak to a human” or “contact support” and the assistant links to your workspace's Contact support page. Submit the message there to send it to our support team; a person replies by email. The link itself does not send a request and uses no AI allowance. You can also open Help → Contact support directly. Product how-to questions use the public documentation; the assistant should cite the matching docs and offer support if the manual does not cover it.

Add contacts to the pipeline

When you ask for contacts to be added to “the pipeline”, the assistant uses the workspace's default pipeline and its first stage. You can name a particular pipeline or stage instead. It looks up those names within this workspace and refuses unknown pipelines or stages. Creating a contact and opening its deal are separate actions; check the tool trail to confirm both were completed.

What the assistant can see

The assistant knows the current workspace, its timezone and currency, your role, and sometimes the page or record you are viewing. It can use permitted tools to look up contacts, opportunities, conversations, bookings, forms, social work, sites, automations, and analytics. It cannot read another client workspace through your current chat.

Under Settings → AI assistant, administrators fill in the Brand kit: what the business does, who it serves, how it sounds, how its images look, and up to six brand images such as products, premises or past work. The logo and brand colour come from the workspace's branding settings. The brand kit is supplied to the assistant, the write-and-improve controls and social caption generation on every request, so drafts are about this business rather than any business in its trade. Each client workspace has its own brand kit.

The same page holds standing instructions, and lists the tools available to the assistant and which ones make changes.

How actions are controlled

  • Every tool is checked against the current member's workspace permissions.
  • Turning off Let the assistant make changes makes its workspace tools read-only; it can still answer questions and draft copy.
  • When actions are on, it can make only supported changes that you ask for. Its tool trail shows what it did, and write actions are recorded in the audit log.
  • It does not publish social posts or forms, activate automations, or delete records. Sending a customer message requires an explicit request in the conversation.

Get useful drafts and follow-up work

Ask for a concise email draft, a form for a particular offer, a pipeline summary, or a social post draft. Give the audience, purpose, constraints, and tone. For example: “Draft a follow-up for the contacts who asked for a quote this week. Keep it under 120 words and do not send it.”

The inline write-and-improve controls use the workspace's configured provider too. AI Studio also has an Images & video area, which runs on the included AI usage or on your own OpenAI or OpenRouter key. Learn about providers in Bringing your own key.

Models, usage, and limits

Use the model picker beside the chat composer to change the model for the next message, even in an existing conversation. Larger models may take longer and use more of your allowance. A single request may require several model calls when it searches or updates records, so usage depends on the work performed, not just the number of messages you sent.

If a request stops at its tool-step limit, ask the assistant to continue with the remaining part. For the current model choices and metering, see Models and Usage and limits.