Skip to content

AI Gateway Playground

Varity TeamCore Contributors

The Playground is the first tab of the AI Gateway page in the Developer Portal. It sends real requests through the AI Gateway, so they are charged to your gateway balance and count against the same account rate limits as your API keys.

The composer has two modes:

  • Build (Beta): build and change your app with a verified Build model; new Build chats start on DeepSeek V4.1 Flash, and the model picker lets you choose another available Build model.
  • Chat: ask, plan, and read with any model. Nothing is written or deployed. The model picker lists your favorites, your recent models, and a short list, then All models for the rest of the live catalog.

A tool the selected model cannot use is shown as unavailable, and its tooltip says why.

Add

  • Files and Folder: attach files, or a whole folder, to your message. A document stays on the message it was sent with, and each message holds up to the selected model’s per-message document limit (capabilities.max_documents_per_message). The × on a sent document removes it from that message. On a model with Varity tools, a dropped folder’s PDFs join the chat’s files as text.
  • Create image: choose an image model and open its workspace.

Tools

  • Web search: search the web and cite sources.
  • Read links: read the pages linked in your message.
  • Deep research (Chat): several private web searches, then one cited answer.
  • Compare models (Chat): send one prompt to 2 or 3 models side by side.

Connect

  • Repository: choose the GitHub repository the assistant works on.
  • Connectors: add remote MCP servers. A connector tool that is not marked read-only asks you to Run or Don’t run before it executes.

Varity tools let the assistant work with your deployments, repository, and builds. They are on whenever the selected model supports streamed tool calls; there is no switch. With them, the assistant can:

  • list your deployments and read their status and logs;
  • list, search, and read files in the connected repository;
  • search and read these docs;
  • get a price quote and plan a new deployment;
  • plan a redeploy, an environment change, or a deletion;
  • propose a code change as a pull request, read a pull request and its CI status, and plan its merge;
  • plan deploying a commit.

Anything that changes your account is a plan first. You approve each step, such as the commit and then the deploy, before it runs. Chat keeps only the reading tools: it writes no files and proposes no deploy, commit, or merge.

Varity tools add input tokens to every message.

After each Build reply, the chat lists the files the reply wrote, or says No files changed in this reply. When the preview shows an error, Fix this error sends it back to the chat; nothing is fixed automatically.

Open the side panel for three views:

  • Preview: run the in-progress app as its files change. After a deployment, Live site switches between that preview and the deployed app. A new deployment opens the live site, and the next file change returns to the in-progress preview.
  • Code: browse the chat’s workspace and connected repository, then open a file in the editor.
  • Logs: follow a deployment’s runtime logs.

View code shows the exact request the Playground would send for the current message, so you can copy it into your own app.

  • History and projects: conversations are grouped into projects and kept on your device, not on Varity’s servers. The AI Gateway stores no prompt or response content.
  • Projects: a new chat in a project shows the project’s Instructions, Files, and Memory. Deleting a project deletes its chats.
  • Project files: open a project’s settings and use Add files or Add folder. A project holds text, code, and SVG files, and PDFs converted to text; Word, PowerPoint, and Excel files are not added to projects. Every chat in the project reads these files with any model that supports Varity tools; in Chat, a model without them gets the files as context while they fit. Each project holds up to 5 MB of files, stored on your device.
  • Memory: notes you save on your device. A note added inside a project applies only to that project’s chats; deleting the project deletes it. While memory is on, notes are sent with each message, and you can see them in View code.
  • Incognito: an incognito chat is not saved to history and does not use memory. It ends when you leave it.