Running headless browser jobs on Cloudflare means a lot of copy-pasting JSON payloads into a chat window every time you want a second opinion. HotBot’s native Cloudflare Browser Rendering AI integration removes that step for signed-in members. Connect the service directly and the assistant reads your rendering data, then helps you review code, write queries, triage incidents, and draft docs.
Set it up once and the screenshot-and-JSON shuffle between tabs is done. The rest of this guide covers the setup and five workflows worth building around it.
What You’ll Need
- A HotBot membership — connectors are a members-only feature. Sign in or subscribe here.
- A Cloudflare account with Browser Rendering (now called Browser Run) enabled. Per the Cloudflare Browser Run docs, you’ll want an API token with
Browser Rendering - Editpermissions for most workflows. - Something real to work on — a PR to review, a flaky crawl job, an incident log.
- No local browser stack required. A hosted browser service is the whole point.
Estimated time: 10–15 minutes for the first connection, then seconds per task. Difficulty level: Beginner-friendly. If you can sign in and paste a prompt, you’re qualified.
Step 1: Sign In and Connect Cloudflare Browser Rendering Inside HotBot
Membership comes first. Connectors don’t exist on the free tier for a practical reason: the assistant works with your live data, so it lives behind the login.
Go to the HotBot login page and sign in, or subscribe if you haven’t yet. Once you’re in, connect Cloudflare Browser Rendering from inside HotBot. The connect flow lives in the app and walks you through it.
After it’s connected, HotBot can reference your Browser Rendering data directly in a conversation. Everything below assumes you’ve done this step.
Step 2: Review Code and PRs
Cloudflare runs up to seven specialized AI reviewers on every merge request — security, performance, code quality, docs — and across 131,246 reviews in a month, their median review took 3 minutes 39 seconds. You don’t need that whole orchestration to get value. A decent model and your diff will do.
If your Browser Rendering work involves Puppeteer or Playwright scripts (the Cloudflare docs note you install Cloudflare’s fork of Puppeteer for Worker-based sessions), HotBot can read the rendered output and the code that produced it. That combination is what catches the “why is this screenshot blank” bugs.
Example prompt:
“Review this Worker that takes a screenshot via the browser binding. Flag any missing
waitUntilorwaitForSelectorhandling that could cause blank captures, and check for hardcoded secrets.”
Model tier: Use HotBot Chat for quick diff reviews and style nits. For a full PR with cross-file logic, step up to HotBot Chat Pro — the 1M-token context means you can drop the whole diff plus related files and it won’t lose the thread.
Step 3: Write and Explain Queries and Extraction Prompts
Browser Rendering’s /json endpoint extracts structured data using natural-language prompts or a JSON schema. Per the Cloudflare /json docs, it handles normalizing article metadata, extracting product info, or converting unstructured pages into typed JSON for downstream pipelines. Writing a clean schema and a prompt that returns exactly what you want is fiddly work.
That’s a good HotBot job. Describe the page, describe the shape you want, and let it draft both the prompt and the response_format schema.
Example prompt:
“I’m hitting the Browser Rendering
/jsonendpoint on a jobs listing page. Write me a prompt plus a JSON schema that returns an array of jobs with title, company, location, and salary. Handle missing salary gracefully.”
Model tier: HotBot Chat Plus works well here — the step-by-step reasoning helps it handle edge cases like nulls and nested arrays instead of glossing over them. If you’re wiring in a specific vendor’s model your pipeline already depends on, a third-party model on HotBot can be the better pick. You get 800+ models under one subscription, so match the model to whatever’s downstream.
Step 4: Triage Incidents and Errors
When a rendering job breaks, it’s usually one of a few causes: a page that never hits network idle, a selector that moved, a timeout. The Cloudflare docs call out the classic fix — set gotoOptions.waitUntil to networkidle0 or networkidle2, or use waitForSelector for faster, targeted waits. Knowing which fix applies to your specific failure is the harder part.
Paste the error, the failing config, and the target behavior. HotBot connects the dots faster than most people can read a stack trace at 2am.
Example prompt:
“This Browser Rendering job times out intermittently on a JS-heavy page. Here’s the config and the error. Walk me through likely causes ranked by probability, and give me the specific
waitUntilorwaitForSelectorchange to test first.”
Model tier: Reach for HotBot Chat Pro. Incident triage rewards deep reasoning — you want the model holding logs, config, and hypotheses at once. The big context window earns its keep here.
Step 5: Draft Docs and Changelogs
Docs are the task everyone puts off. When your extraction pipeline or Worker changes, though, someone has to explain it. HotBot can turn your diff and rendered outputs into a readable changelog or a doc section written in your voice rather than robot-speak.
Example prompt:
“Here’s the before/after of my Browser Rendering crawl config plus two sample outputs. Draft a changelog entry and a short ‘How extraction works’ section for our internal wiki. Keep it plain-English.”
Model tier: HotBot Chat handles most changelog and doc drafts fast. Bump to Chat Plus if you need structured, multi-section documentation that has to flow logically. If you want a diagram or a cover image for the docs, the HotBot Image engine can generate that too.
Step 6 (Bonus): Combine Screenshots + Reasoning
Browser Rendering is built for screenshots and PDFs — a headline use case in the Cloudflare get-started guide. Because Chat Pro supports vision, you can hand it a rendered screenshot and ask why the layout broke, or whether the captured page matches what you expected. Text config and visual output, reasoned about together. Most people don’t realize they want this workflow until they’ve used it.
Without the Connector (Free Tier)
No membership yet? You can still get most of the value with a little manual labor. Paste your Cloudflare Browser Rendering content — the JSON output, the Worker code, the error log — straight into HotBot Chat and prompt away.
It works. You lose the direct, live connection and you’ll be copy-pasting more, but for one-off reviews that’s fine. If you’re doing this daily, the connector pays for itself in saved tab-switching. Check pricing — there’s a free tier, plus $7.95/week or $39.95/quarter.
Limits and Privacy Notes
A few caveats worth stating plainly:
- The connector works with your Browser Rendering data while you’re in a conversation. Treat any secrets — API tokens, credentials — the way you would anywhere, and don’t paste them into prompts unless you have to.
- Cloudflare Browser Rendering has its own usage limits and pricing on Cloudflare’s side. NomadX’s 2026 Cloudflare guide pegs it at roughly $0.09 per browser-hour with a free daily allowance. Your rendering costs are separate from HotBot.
- HotBot is an independent AI chat service. It’s not affiliated with or endorsed by Cloudflare. Third-party names here identify the tools and models you can use, nothing more.
Common Issues
“I don’t see a connect option.” You’re probably on the free tier or not signed in. Connectors are members-only — sign in first.
“My extraction returns junk.” Usually a timing problem. Try the waitUntil: networkidle2 fix from the Cloudflare docs, then ask HotBot to review your prompt and schema pairing (Step 3).
“The model lost track of my big PR.” Switch to Chat Pro’s 1M-token context. Smaller models truncate large diffs; Pro handles them.
What’s Next
Once the connector is running, try chaining workflows — triage an incident (Step 4), fix the config, then have HotBot draft the changelog (Step 5) in the same session. Because the assistant already has the context, each step gets faster.
The full model lineup is worth exploring too. Running the same PR review across a couple of models and comparing what each one catches often turns up gaps you’d otherwise miss.
When you’re ready, sign in and connect it. The 2am incident-triage version of you will thank you.
FAQ
Do I need to be a member to connect Cloudflare Browser Rendering?
Yes. Connectors are a members-only feature — you connect supported apps from inside HotBot once you’re signed in. The free tier can still paste content in manually, but the live connection requires a membership.
Is HotBot an official Cloudflare Browser Rendering product?
No. HotBot is an independent AI chat service and isn’t affiliated with or endorsed by Cloudflare. Third-party names identify the models and apps you can use, nothing more.
Which HotBot model should I use for code review?
HotBot Chat for quick diffs and style checks, HotBot Chat Pro for full PRs with cross-file logic thanks to its 1M-token context. Chat Plus sits in the middle for step-by-step reasoning tasks.
How much does HotBot cost?
There’s a free tier, plus paid plans at $7.95/week or $39.95/quarter. One subscription gets you access to 800+ models plus HotBot’s own Chat, Chat Plus, Chat Pro, and Image engines. See pricing.
Does connecting Cloudflare Browser Rendering cost extra on the Cloudflare side?
Yes, separately. Cloudflare bills Browser Rendering usage on its own — around $0.09 per browser-hour per NomadX’s 2026 guide, with a free daily allowance. That’s independent of your HotBot subscription.
Can HotBot read screenshots from Browser Rendering?
Yes, with HotBot Chat Pro, which supports vision. You can hand it a rendered screenshot alongside your config and ask it to reason about layout issues or verify the capture.
