Skip to content

article

Getting Started with StretchGPT

What StretchGPT is, the key concepts behind it, and the first things to do to hold your first conversation.

StretchGPT is StretchSuite's proprietary AI assistant — think of it as your organization's own ChatGPT, but wired into StretchGroup's products, grounded in your own knowledge, and governed by the controls you already trust. This guide gets you from zero to your first useful answer, and points you to the deeper articles once you're comfortable.

StretchGPT
StretchGPT

What StretchGPT does

At its heart, StretchGPT lets you have a conversation. You type a question or request, StretchGPT replies, and you can keep going — it remembers the recent turns so follow-ups make sense. Three things make it more than a generic chatbot:

  • It knows StretchGroup. Ask about the suite's brands, products, and services and StretchGPT answers from real context.
  • It grounds its answers. When supporting evidence exists in the StretchSearch index, StretchGPT cites its sources inline as [1], [2], so you can verify.
  • It's purpose-built. You can pick an assistant family tuned to the kind of work you're doing, from writing code to research to customer support.

Key concepts

  1. Conversation — a saved thread of messages. Each conversation belongs to you and your organization; nobody else can see it.
  2. Assistant family — the flavor of assistant a conversation uses. The default is Core, a capable all-rounder. Other families (Code, Research, Ops, Support, and more) tune the voice and focus.
  3. Grounding — when StretchGPT retrieves evidence from the StretchSearch index and cites it in the answer.
  4. Activation — whether live model execution is switched on. Until an operator activates the runtime and connects a model provider, StretchGPT runs in a safe, non-live mode and will tell you what's still needed.

Your first conversation

  1. Open StretchGPT from the StretchSuite app launcher (or go to your StretchGPT workspace).
  2. Click New chat. If you want a specialized assistant, choose a family such as Core or Code before you start.
  3. Type your question in the message box and press Send. Keep individual messages under 4,000 characters.
  4. Read the reply. If you see bracketed markers like [1] at the end of a sentence, scroll to the sources list to see exactly what StretchGPT relied on.
  5. Ask a follow-up. StretchGPT keeps the last several turns in context, so you don't need to repeat yourself.
  6. Your conversation is saved automatically. Find it later in the conversation list on the left.

A realistic example

Suppose you're onboarding to a StretchGroup product. In a Core conversation you type:

"Summarize what StretchDrive is and who it's for, in three bullet points."

StretchGPT replies with a tight three-bullet summary. Because the answer draws on indexed StretchGroup material, it ends with a source marker. You then follow up:

"Now write a two-sentence intro I could paste into a welcome email."

Because the prior turn is in context, StretchGPT knows "it" means StretchDrive and produces the intro without you re-explaining.

Understanding activation states

If your workspace hasn't been fully activated yet, StretchGPT still accepts your message and saves the conversation, but instead of a live answer it explains what's missing — for example, that a model provider needs to be connected or that live execution hasn't been switched on. This is by design: StretchGPT is secure by default and never calls a model until an operator explicitly turns it on. See Activating Live Chat for the operator steps.

Tips

  • Lead with what you want. StretchGPT is tuned to answer directly, so a clear ask ("Draft…", "Compare…", "List the steps to…") gets a better result than a vague one.
  • Pick the family that matches the task. A coding question in the Code family gets a more focused answer than the same question in Core.
  • Use the citations. If an answer is grounded, the sources are right there — check them before you rely on a fact.
  • Rate answers. A thumbs-up or a thumbs-down with a correction genuinely improves StretchGPT's own models over time.
  • Keep messages under 4,000 characters. For long documents, paste the key section rather than the whole thing.

Where to go next

  • Understanding Assistant Families — pick the right family for the job.
  • Chatting with StretchGPT — the full conversation workflow.
  • Grounded Answers and Citations — how evidence and [n] citations work.
  • Administrators: Connecting a Model Provider and The Safety Envelope and Execution Gate.

Was this helpful?

Help us improve this article

Use these controls to share whether this answer solved the issue. Feedback helps prioritize updates to StretchSuite Support.