HashAgent
An open-source project that encodes AI agent configurations into URLs and runs them locally in the browser via WebGPU, enabling zero-deployment-cost distribution of AI agents with all data processed entirely on the user's local device.
Timeline (last 90 days)
HashAgent is an AI distribution model that encodes an entire AI agent into a URL and runs it locally in the browser via WebGPU.
The HashAgent project recently appeared on Hacker News.
What HashAgent encodes into the URL is the agent's configuration and logic definition (such as model references, system prompts, tool-calling logic, and runtime parameters), not the large model weights themselves.
Browser-side AI execution is constrained by device VRAM and compute power; currently, WebGPU can smoothly handle mostly small models on the order of billions of parameters, making it suitable for lightweight vertical tasks rather than general complex reasoning.
Running AI locally requires downloading model weight files of hundreds of MB or even several GB on the first use, causing noticeable latency that can be mitigated through browser caching.
HashAgent received only 5 points and 0 comments on Hacker News, indicating it is at a very early stage.
HashAgent's local execution mode shifts inference costs from the sharer to the user side, requiring the sharer only to generate a link at virtually zero cost and with no risk of high server concurrency.
All Facts (5)
HashAgent is an AI distribution model that encodes an entire AI agent into a URL and runs it locally in the browser via WebGPU.
50%UnverifiedWhat HashAgent encodes into the URL is the agent's configuration and logic definition (such as model references, system prompts, tool-calling logic, and runtime parameters), not the large model weights themselves.
50%UnverifiedBrowser-side AI execution is constrained by device VRAM and compute power; currently, WebGPU can smoothly handle mostly small models on the order of billions of parameters, making it suitable for lightweight vertical tasks rather than general complex reasoning.
50%UnverifiedRunning AI locally requires downloading model weight files of hundreds of MB or even several GB on the first use, causing noticeable latency that can be mitigated through browser caching.
50%UnverifiedHashAgent's local execution mode shifts inference costs from the sharer to the user side, requiring the sharer only to generate a link at virtually zero cost and with no risk of high server concurrency.
50%