> For the complete documentation index, see [llms.txt](https://docs.robincompute.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.robincompute.org/providers/browser-worker.md). # Browser Worker A browser worker turns a Chrome tab into a paid inference node. You install nothing: connect a wallet on the Earn tab, pick a model, and jobs start routing to your GPU. Inside the tab, inference runs through the WebLLM runtime on WebGPU. Prompts execute on your own hardware, in the browser, and tokens stream back to whichever client submitted the job. *** ## What you need | Requirement | Details | | ----------- | ---------------------------------------------------------------------------------------------- | | Browser | Chrome 113 or newer, or any Chromium browser with WebGPU turned on | | GPU | Any card the browser can drive over WebGPU, with at least 3GB of GPU memory exposed to the tab | | Wallet | An Ethereum wallet: MetaMask, Rabby, or Robinhood Wallet | | Network | A reliable connection. Streaming quality tracks your upload bandwidth. | | Stake | None. Browser workers never stake. | **Checking for WebGPU:** open `chrome://gpu` and find the "WebGPU" entry in the graphics feature status list. "Hardware accelerated" means you qualify. **What will not work:** Safari (its WebGPU implementation is unfinished), Firefox (WebGPU ships disabled in most builds), and every mobile browser. Worker mode is desktop Chromium only. *** ## Setup **1. Connect a wallet** Visit `robincompute.org/app` and open "Earn" from the navigation. Connect an Ethereum wallet when prompted (MetaMask, Rabby, or Robinhood Wallet). USDG payouts land at whichever address you connect here. **2. Pass the compatibility check** The moment you connect, the Earn tab probes your setup. It determines: * whether WebGPU is available in your browser * how much GPU memory the browser can see * which models fit inside that VRAM budget If WebGPU is missing, you get an error message with instructions for enabling it, or a pointer to a supported browser. **3. Pick a model** You only see models that fit your detected VRAM. Bigger models pay more per job; the trade is that the rest of your machine feels sluggish while one runs. | Model | Tier | VRAM required | Earnings per job | | ----------------- | -------- | ------------- | ---------------- | | Qwen3 1.7B (Q4) | Lite | 3GB | $0.015 | | Llama 3.2 3B (Q4) | Lite | 4GB | $0.015 | | Qwen3 8B (Q4) | Standard | 6GB | $0.060 | | Mistral 7B (Q4) | Standard | 6GB | $0.060 | **4. Go live** Hit "Start Earning" and the page pulls the model into browser memory. Expect 15 seconds to 2 minutes depending on model size and connection speed. The weights stay in browser storage after the first load, so later starts are fast. Once the model is loaded, your worker registers with the orchestrator mesh and jobs arrive on their own. The page shows each incoming job and a running USDG total. Your part is done. *** ## What happens during a job Each job that reaches your tab goes through six stages: 1. The orchestrator delivers an encrypted payload over a WebSocket connection. 2. A Web Worker thread decrypts it with the ephemeral session key. 3. WebLLM runs the prompt on your GPU through the WebGPU API. 4. Tokens stream back to the orchestrator as the model produces them. 5. When generation finishes, the worker hashes the complete output with SHA-256, signs the hash with the wallet key, and submits that proof to the `settlement` contract. 6. Settlement pays USDG to your wallet, usually confirming in under a second. RobinCompute sponsors the gas through ERC-4337, so you never touch ETH. Inference lives on a Web Worker thread, which keeps the tab responsive and lets you browse elsewhere while earning. Heavy GPU work in another tab will cut into your inference throughput, though. *** ## Getting paid USDG hits your connected wallet after every completed job, each one a settlement transaction you can look up. No minimum balance to withdraw, no batched payouts. The browser tier pays 75% of each job's value. You never need to stake to run a browser worker, but the 85% rate stays out of reach on this tier: it belongs to `robincompute-node` operators holding an active $RCOMPUTE stake. During the beta, browser workers also collect a 2x $RCOMPUTE reward multiplier on top of their USDG. Those $RCOMPUTE rewards come out of the Community/Provider rewards pool and vest over 18 months. *** ## Stopping and coming back Close the Earn tab and your worker goes offline immediately. A job caught mid-flight times out and reroutes to another worker; the portion you completed before closing goes unpaid. Nothing else is lost. Your USDG balance and reputation live on the chain, not in the tab. Reconnect your wallet later and your full history is right where you left it. *** ## Troubleshooting **"WebGPU not available"** Turn on hardware acceleration at `chrome://settings/system`. Already on? Update your GPU driver. **The model will not load** The browser may see less GPU memory than the model needs. Pick a smaller model, or free up VRAM by closing tabs running video, 3D content, or other WebGPU apps. **No jobs are coming in** Freshly loaded workers can take up to 60 seconds to appear in the orchestrator mesh. Still nothing after several minutes? Check your connection, then confirm the dashboard lists your worker as "Active." **Hourly earnings feel low** Lite-tier models pay less per job than the larger models native workers run. That is the tier, not you. If you own an NVIDIA card or an Apple Silicon machine, the \[Native Worker]\( ) tier earns considerably more per hour. --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.robincompute.org/providers/browser-worker.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.