REA 6.4 brings reverse engineering to coding agents; Copilot for JetBrains gets admin controls

REA is an MIT-licensed tool that plugs into coding agents such as Claude Code, Codex, Cursor and Gemini CLI as an MCP server (the standard way agents call external tools). Once connected, agents can analyze native binaries, Electron apps, .NET assemblies and websites without source code. A single command, npx rea-agents setup, registers it with your agent, and it drives existing disassemblers (tools that turn machine code into readable code) like Hopper, Ghidra or IDA. Version 6.4 adds Android device inspection and APK extraction over ADB, JEB decompiler integration, and summary views for very large analyses.

It widens what you can hand to an agent: security reviews, making sense of legacy binaries, or digging into undocumented internal tools. Every conclusion comes with its evidence and limits, so results are easier to check. Because it's built to inspect other people's apps, check license terms and local law on reverse engineering before you point it at anything.

This is an update to the GitHub Copilot plugin for JetBrains IDEs. Enterprise admins can now use managed settings to pick the default agent model for new conversations. A Fix action in the diagnostics menu opens inline chat and repairs the issue in agent mode. A new setting stops Copilot and Claude MCP servers from starting automatically, and there are smaller touches too, like copying and editing your own chat messages.

Teams that want to keep cost and quality consistent need control over the default model, and JetBrains users now have that lever too. Turning off MCP autostart makes it easier to keep unreviewed tools from spinning up on their own. Support for JetBrains IDE 2025.1 has ended, so check which IDE versions your org runs.

A Mac meeting-notes app from Google's AI Edge team. It transcribes mic and system audio on the device (speech-to-text) and uses the transcript to turn your rough shorthand into proper notes. Point it at PDFs, Google Docs or Office files and it builds a local knowledge base. When questions come up mid-meeting, it shows answers with citations. Retrieval runs on EmbeddingGemma 2 (a 740M-parameter embedding model that puts text, images, video and audio in one vector space), and chat runs on Gemma 4.

It's Google's reference build for bundling on-device speech-to-text with RAG (pulling your own documents into answers) in one app. That matters for teams whose security rules rule out cloud note-takers. Its offline, local-first design sets it apart from cloud services like Granola. If you're building on-device voice or search features, it's a useful benchmark for what the Gemma family can do.