PLVA
PLVA is a private, teachable computer-use agent.
Core trick
Replaces detected sensitive text in screenshots with local OCR tokens before sending frames to the model, and restores real values only in approved local typing actions.
Description
PLVA is a private, teachable computer-use agent. It helps people delegate browser work while keeping detected sensitive values on their own machine. Local OCR replaces private text in screenshots with tokens before GPT-6 Astra sees the page; the runtime restores values only in approved local typing actions. An outgoing-request inspector shows exactly which protected frames and actions were sent.
Our working Windows browser prototype supports real Astra execution, local token insertion, and independently checked privacy tests. A separate skill-learning prototype converts protected recordings into reviewable procedures and prepares them for new inputs. Connecting that pipeline to live execution is the next step toward our goal: do a task once, learn the procedure, and reuse it safely. The video is a scripted product walkthrough; live-runtime evidence and current limits are documented in the repository.
README excerpt
# PLVA — Private AI Assistance for Client Work **Protecting client information is the central purpose of PLVA.** The project lets an AI assistant help with browser workflows while keeping detected private details behind a local privacy boundary. The assistant receives protected screenshots and opaque references; permitted uses of the underlying client values happen locally, under checks enforced by PLVA. Client names, email addresses, phone numbers, shipping details, and credentials often appear in the same screens an assistant needs to understand. PLVA separates the information needed to plan a task from private values that should remain under the user's control. Its main contribution is the complete privacy path around AI assistance: local detection, screenshot protection, private references, controlled execution, protection of the next observation, and inspectable evidence of what actually reached the provider. The project brings together three workstreams: | Workstream | Contribution | Current delivery status | | --- | --- | --- | | **PLVA browser privacy + Astra** | Protect observations, retain values locally, enforce permitted actions, and show actual submitted images and provider receipts. | Implemented and running in the integrated Windows browser app; tested on the workflows described below. | | **Private reasoning + NemoClaw** | Add local reasoning over private information, with a separate isolation/deployment boundary and constrained recommendations. | Implemen