Contextberg

Contextberg runs quietly and records three streams at once: screenshots from every window, your browser history, and transcripts from Claude Code, Cursor, and the terminal. You don't tag anything or write notes. The point is that the record exists already, so when you need it, it's there. That's the difference between a memory app and a to-do list you have to maintain.

The app ships with an MCP server, which means any compatible agent can connect without extra wiring. You skip the usual config files and just point the agent at Contextberg. For people who've spent an afternoon hand-building a memory pipeline, this is the shortcut. It's the main reason the tool fits into an existing workflow instead of replacing it.

Contextberg sorts what it captures into activity memory (what you did), daily memory (progress grouped by date), and long-term memory (the tools and patterns you keep reaching for). Each layer answers a different question. Activity memory is for "what was I doing an hour ago." Long-term memory is for "how do I usually set up a new project." The split keeps the agent from drowning in raw logs.

You choose where the reasoning happens. Options include your Codex sign-in, a Gemini or OpenRouter API key, Contextberg Cloud, or a local model through LM Studio. The local route keeps everything on your machine. Every other route sends only the context needed for that request, according to the company. If you switch providers, your memory store stays put.

When you sit back down, Contextberg summarizes what you were doing before you left, pulled from recent activity, browser history, and agent transcripts. You can then dig into any part of it in chat. This sounds small until you've lost twenty minutes rebuilding state after a meeting. The summary is the feature most people will notice first.