Nex Life Logger Skill

What it does

• A background collector logs your productive computer activity into a local SQLite database: browser history from Chrome, Edge, Brave and Firefox, which window you had focused, and the caption text of YouTube videos you watched.
• A CLI lets you, or an agent on your behalf, search all of it with full-text search and filter by date, source, kind and category.
• Extracts the topics, tools and languages you have been spending time on, and can generate daily, weekly or monthly summaries.
• Exports to JSON, CSV or HTML whenever you want your data out.

What it does not log

• Password managers, keychains, vaults, banking windows and private browsing windows are never logged.
• Chat and messaging apps are filtered out.
• Entertainment, politics and news are filtered out; only productive content is kept.

Network access, in full

Your activity data never leaves your machine. Everything is written to a local SQLite database at ~/.life-logger/. There is no telemetry and no analytics. There are exactly three points where this skill opens an outbound connection, and all three are documented in SKILL.md:

1. YouTube caption tracks, during collection. When the collector sees a productive YouTube video in your history it asks YouTube for that video's public captions so the spoken content becomes searchable. Only the 11-character video id is sent. Set LIFELOGGER_FETCH_TRANSCRIPTS=0 and the collector makes no outbound connections at all.
2. An LLM endpoint, only if you configure one, and only for summaries. There is no default endpoint and no bundled key. Summary generation is inert until you set a key and a provider. The provider shortcuts are for endpoints you pick yourself: api.openai.com, api.groq.com, dashscope-us.aliyuncs.com, or Ollama on localhost, which keeps summarisation on your own machine. Every other command works with no provider configured and contacts nothing.
3. PyPI, once, during setup. setup.sh installs openai, psutil and youtube-transcript-api into a virtual environment. Nothing is fetched at runtime.

Your API key, if you set one, is stored locally with restricted permissions and is only ever sent to the base URL you configured.

Use cases

• Asking your agent what you were working on yesterday, last week or last month
• Recovering the name of a tool or article you saw but did not bookmark
• Seeing where your time actually went before you write a timesheet or an invoice
• Searching what was said in a YouTube video you half-watched
• Generating an honest weekly summary of your own work

Known limitations

The collector must be running for data to be captured, so it cannot tell you about time before you installed it. Browser history gives page visits, not time-on-page, so durations are approximations. Transcripts only exist for videos that have captions. AI summaries need an LLM provider you supply yourself. Tested on Windows, macOS and Linux with Python 3.9 or newer.