Kryzz AI: an open-source Android agent that shows its plan before acting

·2 views

The developer of Kryzz AI, an MIT-licensed open-source Android chat app, built it on weekends and evenings over several months. The app supports text chat, voice conversations, and an agent that performs small multi-step tasks. A core design choice is that the agent displays a live plan card and a visible work log, letting users stop it early and check where answers came from. Safeguards include restricted link fetching, a second review pass for longer answers, and a default $0.25 per-task cost cap adjustable up to $2. The app also keeps chats on the device with bring-your-own API keys and no telemetry, though model calls still go to the chosen provider.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Wild boar and deer in the Netherlands adapt their behavior to the return of wolves

Wildlife cameras in the Netherlands have recorded wild boar and deer changing their habits in response to the presence of wolves. The footage suggests the animals are adjusting instinctively to the returning predator. The report is based on a video published by Dutch broadcaster NOS. Details beyond the observed behavioral changes are limited in the available source material.

0 comments Read more at Hacker News

ProgrammingHacker News ·

Could MCP Become the Shared Base for Decentralised Smart Search?

A Hacker News discussion links to an article asking whether the Model Context Protocol (MCP) is emerging as a common foundation for smart, decentralised search. The piece examines how MCP might let AI tools connect to multiple data sources without relying on a single central index. The post has drawn 8 points and 2 comments on Hacker News so far. The source material provided is limited to the link and engagement metrics, so the article's specific arguments are not detailed here.

0 comments Read more at Hacker News

ProgrammingDEV Community ·

Seven Sports Builds Amblyotube, a VR App for Vision Training Inspired by NASA Research

Seven Sports developed Amblyotube, a first-person virtual reality application for Meta Quest aimed at vision training. The team drew on insights from NASA's CRS-35 research mission when designing the app. The core engine applies different processing to each eye, using a blur shader for the dominant eye and an AI sharpener for the lazy eye before compositing the gaze. The resulting image is combined through a binocular fusion step to support joint eye use. The project's code is open source, and the developers invite contributions via a public GitHub repository.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AgentMesh Uses FalkorDB Graph Memory to Help AI Agents Retain Project Context

A developer built AgentMesh during FalkorDB Graph Hacks 2026 as a persistent memory layer for AI agents. The project addresses the difficulty agents have retaining project decisions across separate sessions and handling changed requirements. Instead of storing knowledge as isolated text, AgentMesh models memories, sessions, and decisions as a connected graph so relationships are part of the stored data. Three specialized agents handle extraction, retrieval, and contradiction review, allowing a newer decision that conflicts with an earlier one to be flagged. The developer presents the graph-based approach as a way to retrieve prior context without requiring users to restate earlier decisions.

0 comments Read more at DEV Community