Skip to content

product hub

StretchGPT

StretchSuite's proprietary AI assistant — grounded, multi-turn chat with purpose-built assistant families, plus an operator console for model providers, the training pipeline, GPU worker nodes, and a hard safety envelope.

StretchGPT dashboard
StretchGPT dashboard

StretchGPT

StretchGPT is StretchSuite's answer to ChatGPT — a proprietary AI chat platform backed by the StretchAI control plane. Everyone in your organization can hold natural, multi-turn conversations with an assistant that knows StretchGroup's brands, products, and services, and can help with everyday work as well. Behind the chat window sits a full control plane where operators and admins manage the models that generate answers, the model providers those models run on, a human-reviewed training pipeline, GPU worker nodes, and a hard safety envelope — all from one workspace console.

Who it's for

  • Everyday users who want a fast, accurate assistant inside the suite — drafting, summarizing, answering questions, and reasoning over your StretchGroup context.
  • Team leads and power users who want purpose-built assistant families (Core, Code, Research, Ops, Support, and more) tuned to a specific kind of work.
  • Workspace operators and admins who connect model providers, curate training data, promote models through staging to production, and register the GPU nodes that do the heavy compute.

What you can do

Chat, grounded and persistent. Start a conversation, pick an assistant family, and ask. Every conversation is saved, isolated to your account and organization, and remembers the recent turns so follow-up questions just work. When StretchGPT can find supporting evidence in the StretchSearch index, it grounds its answer and cites the sources inline as [1], [2], so you can check the facts.

Choose the right assistant family. StretchGPT ships nine families — core, code, research, ops, learn, support, media, creative, and vision — each with a voice and focus suited to that kind of work. Core is the default all-rounder.

Rate answers and improve the model. A thumbs-up marks an answer as a good example; a thumbs-down plus a correction captures a better one. Both feed a human-reviewed training queue that makes StretchGPT's own models better over time.

Experiment in the Prompt Playground. Operators can start from a library of ready-made templates (support triage, project status briefs, prompt critique, retrieval-grounding checks), tune parameters like temperature and max tokens, save their own prompts, run them, and compare runs side by side.

Connect and govern model providers. Admins register model backends with their own API keys, set a primary provider and default model, order a fallback chain, and rotate or revoke keys. Keys are encrypted at rest and never returned by the API. StretchGPT prefers your proprietary self-hosted model first, then falls back to a teacher model.

Run a real training pipeline. Draft supervised (SFT) and preference (DPO) training candidates with teacher assistance, review and approve them, export an approved dataset, dispatch training jobs to a GPU node, then evaluate and promote the resulting models through a governed lifecycle where staging and production require a recorded evaluation and a human sign-off.

Register GPU worker nodes safely. Compute nodes authenticate with scoped, single-purpose credentials, poll for work over outbound HTTPS (no inbound listener required), and are protected by a timestamp-and-nonce replay guard.

The safety envelope

StretchGPT is secure by default. Live model execution, worker execution, training, dataset writes, and model writes are gated behind an execution gate that runs in dry-run mode until an operator explicitly activates the runtime. Provider keys are encrypted at rest and never exposed to the browser, raw prompts from the public site are not stored, and every sensitive write (connecting a provider, approving training data, promoting a model) requires step-up admin authorization. Each account's chats are isolated by organization and by user, so one account can never reach another's conversations.

How it fits the suite

StretchGPT is the AI front door for StretchSuite. It grounds answers in the StretchSearch index, closes its learning loop through the StretchAI training pipeline, and its models power AI features across other suite apps. Because it's part of the suite, it shares your single sign-on, your organization tenancy, and the same admin controls you use everywhere else.

Where to start

New here? Begin with Getting Started with StretchGPT, then learn how to choose an assistant family and hold a conversation. If you administer the platform, jump to Connecting a Model Provider, Reviewing the Training Queue, 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.