Web Agent

When you issue a command, the Planner agent orchestrates 20+ specialized sub-agents to complete the task:

• 01

  Action Agent — Handles clicks, typing, and navigation

• 02

  Extraction Agent — Identifies and structures data from the DOM

• 03

  Crawl Agent — Manages pagination and multi-page discovery

• 04

  PDF / File Agent — Reads and fills complex forms, uploads documents

The agent can reference multiple open tabs simultaneously — comparing data across sites, moving information between systems, or processing dozens of URLs in parallel via Sheets Workflows.

Once a multi-step workflow succeeds, you can replay its recorded steps without another planning pass. Shortcuts, Schedules, and Triggers can build on that recording; results still depend on the page and inputs available at run time.

Managed runs use GLM 5.3 Flash on the free plan (in ad-supported Free Mode, with a daily fair-use limit), with DeepSeek Flash, DeepSeek Pro and Gemini Flash also selectable on paid plans. Gemini Pro has been retired as a managed option — saved tasks that still request it run on Gemini Flash automatically. Pick your model from Model Settings in the composer.

Save credits by bringing your own key: the "Add your LLM API" flow accepts keys from any major provider — OpenAI GPT models, Google Gemini, Meta Model API (Muse Spark), DeepSeek, OpenRouter models (MiniMax, Hunyuan, and more), or your own local/custom endpoint — and you can also sign in with your ChatGPT plan directly. Keys are stored encrypted and work in cloud runs too. For Gemini specifically, typing /add-gemini-key in the extension walks you through getting a free key from Google AI Studio. If your own provider errors mid-task, the run falls back to a platform model so it completes — covered by Free Mode on the free plan while today's allowance lasts, billed in credits otherwise. Your key runs the model, so nothing deducts from the free plan's daily Free Mode allowance.

Every task carries a credit ceiling — a cap on what one run may spend before it stops to check with you. It exists because a task that loops, crawls further than expected, or gets stuck retrying can otherwise burn through credits unattended. The default is 100 credits on paid plans and 25 on the Free plan; the typical task finishes in well under 15.

When a run reaches the ceiling it pauses and shows what it has spent, what it produced so far, and what is still outstanding. Answer "Continue" and it resumes from exactly where it stopped with a fresh budget — nothing already done is repeated — or "Stop here" to keep the result and end the task. Scheduled runs are the exception: nobody is watching a cron run, so they stop at the ceiling and report their partial work instead of waiting forever for an answer.

Raising or turning off the ceiling

The ceiling is yours to set. If your work genuinely costs 200-400 credits a run, raise it once and the agent stops interrupting you:

• 01

  Extension — Settings (the gear at the bottom of the side panel) → Agent Settings → "Credit Ceiling / Task".

• 02

  Cloud — Settings → Storage & Execution → Credit Ceiling / Task.

• 03

  API — options.limits.creditCeiling on a single execute call, which overrides the saved setting for that run. See Agent API.

Setting the value to 0 turns the check off entirely. Blank (cloud) or the extension's disabled state falls back to the plan default. Each surface keeps its own value — the extension's setting governs extension runs, the cloud setting governs cloud and API runs.

Drag and drop files into the chat or attach them via the toolbar. The agent has specialized capabilities for PDFs: reading content, filling form fields, and generating new documents. It can also upload attached files to web pages when prompted (e.g., "fill this job application and upload my resume").

Where downloaded files go

When the agent saves a file — a PDF it opened, or a document it generated and you asked it to download — it uses Chrome's normal download flow, so the file lands in your browser's Downloads folder. The run's result lists each file (Downloaded: name.pdf) and shows a 💾 chip; click it, or open chrome://downloads in a new tab, to see every file with a "Show in folder" link.

• 01

  Extension runs: files are on your computer, in Chrome's Downloads folder — chrome://downloads is the quickest way to find them.

• 02

  Cloud runs: the browser isn't yours, so downloaded files are backed up to your Files and attached to the run as file artifacts you can open or download.

• 03

  Generated PDFs and webpages are stored for you and linked from the result (PDF: / Webpage:); say "and download it" to also save a local copy.

Everything in your chat — messages, files, images — becomes context for the AI. Keeping context clean leads to better results.

• 01

  Start fresh chats for new tasks — stale context can confuse the planner

• 02

  Only enable the MCP/custom tools you actually need for the current task

• 03

  Use Personal Context in settings to store persistent info (resume, company details) the agent remembers across all chats

• 04

  Click the ✨ Enhance button (Cmd/Ctrl+P) to have a specialized model rewrite your prompt for clarity before sending

The Chrome Extension is where you build and test workflows. Export successful workflows to rtrvr.ai Cloud for 24/7 headless execution — no laptop required. Recordings, tools, and KB context travel with the workflow automatically.

• 01

  Export any workflow from the extension to Cloud with one click

• 02

  Scale to thousands of parallel URLs via the API

• 03

  Connect your account to WhatsApp — send text or voice messages to trigger automations and receive results on your phone

• 04

  Share your best automations via link — recipients import with one click, including recordings, tools, and all context