Settings tour

Settings is where you connect the providers LoopCodeLab uses on your behalf. Agents build on your account, so you either paste an API key or sign in with a subscription, and LoopCodeLab stores what you give it encrypted. You only need to connect one coding agent to start, and everything else is optional and only matters for the kinds of builds you actually run.

The page is split into tabs across the top. A Find a provider search box sits next to the tabs, so if you know the name you want you can jump straight to its card. Every card carries the same set of buttons (Test, Reveal, Track, and Remove), covered at the end of this page.

Subscriptions

The first tab is a planning dashboard rather than a place to paste keys. For any provider you have tracked, it shows days left until renewal, a usage bar, and a pace note so you can see at a glance whether you are on track to run out. You fill this in yourself with the Track button on a provider card. Nothing here is a secret, and none of it ever gates a build. See Subscriptions for more.

Agents

This is the important tab. These are the coding agents that plan and build your projects, plus your optional GitHub token. Connect at least one agent. Each card lets you authorize one of a few ways, and any single method is enough.

• Claude is the most reliable builder. Connect an Anthropic API key, your Claude Pro or Max sign-in, or a flat-rate coding plan.
• Codex is OpenAI’s coding agent. Connect an OpenAI API key or your ChatGPT plan sign-in.
• Antigravity is Google’s coding agent and runs on your Antigravity Pro plan. Sign in with your Google account, no key needed.
• Qwen is Alibaba’s agent. Connect a DashScope API key or the free qwen.ai sign-in.
• Kimi is Moonshot’s Kimi Code CLI. Connect a Moonshot API key or your Kimi subscription.
• Grok is xAI’s Grok Build CLI. Connect an xAI API key or your SuperGrok or X Premium+ sign-in. The subscription sign-in also unlocks Grok Imagine for build media.
• Vibe is Mistral’s Vibe Code CLI. Connect a Mistral API key.
• GLM (BytePlus Coding Plan) is a budget worker that works out of the box on the platform plan. Add your own BytePlus ARK key to run on yours. One ARK key also covers Seedance video, so a key saved here or on the Video card is reused for both.
• NVIDIA builds stories on NVIDIA-hosted models on a free tier. It is a worker, so it needs a Claude, Codex, or Kimi agent as the reviewer. The same key also powers your chat and planning.
• GitHub is optional. A token lets finished builds push to your own GitHub account. The token must be able to create repositories, or the build ends with a push failure.

At the bottom of this tab is an Agent fallback order card. If an agent runs out of quota mid-build, LoopCodeLab fails over to the next connected agent in these chains and keeps the story’s committed progress. You can reorder the master (reviewer) chain and the worker (builder) chain here.

For how to get each of these keys and sign-ins, see Coding agent keys, and for the token in particular, GitHub token.

Media

Cards on this tab generate images, video, music, and voiceover for your builds, or pull free stock footage for B-roll. Images use your token-plan or Qwen key, which is already connected once you set up Qwen. Video, music, and voiceover each need their own key below and are off by default per build, so you turn them on and set a cap when you set the build up.

• Stock media (Pexels) searches free stock photos and video without spending generation credits.
• Video (BytePlus ModelArk) generates short hero and product-demo videos. This uses the same ARK key as the GLM agent, or you can use Grok Imagine on a Grok subscription instead.
• Music (Suno) generates background music and sound.
• Voiceover (ElevenLabs) generates narration. The Grok sign-in can also cover voiceover.

For how to get these keys, see Media & research keys.

Research and data

These cards ground your builds in real information and power the live editing sandbox. Builds work fine without any of them, so connect them when you want the extra grounding.

• Perplexity grounds your build plans in current web facts, so plans cite real competitor features and live details instead of guessing from training data.
• Apify pulls real datasets into a build, so workers can seed with live data instead of placeholder content.
• Live edit sandbox (Daytona) powers the “Go live” button in Visual Edit: your finished app runs hot-reloading in an isolated cloud sandbox while you edit. Without a key, editing still works using the static preview.

For how to get the Perplexity and Apify keys, see Media & research keys.

Apps and stores

This tab holds two groups of credentials that only matter when you are shipping to an app store. Both are empty by default, and you add your own when you need them.

Mobile app and backend is for Flutter app builds:

• Firebase is an optional backend for accounts, cloud database, and push. The easy way is to sign in with the Firebase CLI so builds auto-configure, or you can paste your own google-services.json.
• Google Play takes a service account JSON so the “Submit to Play” step can upload releases through a CI workflow.
• Codemagic (iOS) takes an API token for the cloud-macOS CI that builds and submits your iOS app to TestFlight.

See Google Play & Firebase and Apple & iOS.

Windows desktop and Store is for web-app builds you package for Windows:

• Microsoft Store takes the three Product-identity values from Partner Center, so the “Build Store package” step can use them without you retyping them each build.
• Windows signing (installer) takes a certificate JSON to code-sign the direct-download installer for clean sideloading. The Store does not need this, since Microsoft re-signs Store uploads.

See Windows desktop & Store.

Living docs

These cards connect an optional provider that powers the “Living docs” button on a finished build, which keeps a project’s documentation up to date on GitHub Actions. OpenAI and Anthropic keys from the Agents tab also work for this, so these are extra options rather than a requirement.

• OpenRouter reaches many models through one key.
• Baseten runs GLM or Kimi on an OpenAI-compatible endpoint.
• Fireworks is another OpenAI-compatible option for the same job.

See Living docs for how the feature works.

MCP and plan

The last tab holds two things: the MCP servers you can wire into builds for tools like Google Docs, Sheets, Drive, Gmail, and Calendar, and a note of the plan your account is on.

The buttons on every card

Once a credential is saved, each provider card gives you a few controls:

• Test runs a quick check that the saved credential is still valid. It is a cheap sign-in probe, so it spends nothing, and it lets you confirm a key works before you rely on it in a build.
• Reveal shows a saved key on screen for ten seconds, with a Copy button, then wipes it. It only appears for plain API keys, never for a subscription or JSON sign-in, and an admin can turn it off for the whole deployment.
• Track opens a small dialog for your planning notes about that subscription: start and end dates, peak and off-peak hours, current usage, price, a dashboard link, and free-form notes. This is display only. It never holds a secret and never gates a build, and it feeds the Subscriptions tab.
• Remove deletes the credential. The provider stops working until you connect a new one.

Two ways to connect

Most cards accept either of two methods, and either one is enough on its own:

• API key. You paste a key from the provider and press Save. This bills pay-per-token, or a flat rate for a coding plan.
• Sign in. You press the sign-in button and a short login flow opens in your workspace. You open the link it shows, authorize on the provider’s own page, and paste back a code if it asks for one. This runs builds on your subscription, and for some providers there is nothing to paste at all. A saved sign-in shows as connected once the credential file exists in your workspace.