The Radar
Saturday, 25 April 2026
Today's picks
Vision Banana
AI PlatformsInstruction-tuned image generator that beats SAM 3 on segmentation and Depth Anything V3 on metric depth estimation.
Google DeepMind's latest proves that image generation pretraining might be to computer vision what GPT was to language. The benchmark numbers against established models like SAM 3 suggest we're watching the emergence of unified vision models that can handle multiple tasks without specialised training.
GitNexus
Developer ToolsOpen-source MCP-native knowledge graph engine that gives Claude Code and Cursor full codebase structural awareness.
Finally, someone tackled the fundamental problem of AI coding assistants editing code they don't actually understand. GitNexus builds knowledge graphs of your entire codebase structure, giving agents proper context instead of blind pattern matching. With 19,000 GitHub stars already, developers clearly needed this.
Also on the radar
DeepSeek-V4
AI PlatformsDeepSeek's new compression techniques make million-token contexts actually practical rather than just technically possible. The dual model approach with V4-Pro and V4-Flash shows they understand inference economics matter as much as raw capability.
Browser Harness
AI AgentsBrowser automation for LLMs that actually works across different sites without custom scripting. The freedom to interact with any web interface could unlock proper web agents instead of the limited API-only tools we have now.
AI Security Governance Framework
AI SecurityFinally, a practical playbook for AI governance that goes beyond compliance theatre. Mend's framework tackles the real operational challenges of securing AI systems before the next security incident forces the conversation.
Stash
AI InfrastructureThe missing infrastructure piece for building stateful AI agents. Instead of reinventing memory systems for every project, Stash provides the persistent context layer that makes agents actually remember things between conversations.
Hacker News
Show HN: Browser Harness – Gives LLM freedom to complete any browser task
97 pts 52 commentsAn open-source tool that lets LLMs interact with web browsers to complete tasks automatically. The project aims to give AI agents the ability to navigate and use any website without requiring custom scripting or API access.
Open source memory layer so any AI agent can do what Claude.ai and ChatGPT do
25 pts 6 commentsStash provides persistent memory capabilities for AI agents, allowing them to maintain context across conversations. This addresses the fundamental limitation of stateless AI interactions by providing a shared memory infrastructure.
MenteDB – open-source memory database for AI agents (Rust)
16 pts 1 commentsA Rust-based memory database designed specifically for AI agents to store and retrieve contextual information. The project focuses on providing efficient memory management for agent-based AI systems.
AI Agent Designs a RISC-V CPU Core from Scratch
9 pts 0 commentsAn AI system successfully designed a complete RISC-V processor core without human intervention. The achievement demonstrates AI's capability to handle complex hardware design tasks that traditionally required significant human expertise.
Show HN: VT Code – Rust TUI coding agent with multi-provider support
6 pts 2 commentsA terminal-based coding assistant built in Rust that supports multiple AI providers. The tool provides a text user interface for developers to interact with various AI coding models directly from their terminal.