I built ShellScribe because I kept forgetting what I actually did yesterday beyond a wall of shell history and vague git commits.
ShellScribe runs as a local first shell hook that quietly logs every command you run with exit codes timestamps and working directory context.
It automatically detects your current git repository and tags sessions by project so you can later see exactly what you worked on and when across all your codebases.
On top of the raw data it uses LiteLLM to generate short plain English summaries of a session giving you a git log for your whole dev life that actually reads like a human wrote it.
There are built in commands for AI recaps fuzzy searching your entire command history exporting markdown reports for daily standups and viewing productivity stats like top commands and active hours.
Everything is fully local first with zero telemetry supports many AI models via LiteLLM and is released under the MIT license.
Installation is cloning the repository running the installer reloading your shell and starting your first shellscribe start session which usually takes under one minute.
I would love feedback from heavy terminal users on what it would take for this to become your default memory of what you have been building breaking and shipping.
I built ShellScribe because I kept forgetting what I actually did yesterday beyond a wall of shell history and vague git commits. ShellScribe runs as a local first shell hook that quietly logs every command you run with exit codes timestamps and working directory context. It automatically detects your current git repository and tags sessions by project so you can later see exactly what you worked on and when across all your codebases. On top of the raw data it uses LiteLLM to generate short plain English summaries of a session giving you a git log for your whole dev life that actually reads like a human wrote it. There are built in commands for AI recaps fuzzy searching your entire command history exporting markdown reports for daily standups and viewing productivity stats like top commands and active hours. Everything is fully local first with zero telemetry supports many AI models via LiteLLM and is released under the MIT license. Installation is cloning the repository running the installer reloading your shell and starting your first shellscribe start session which usually takes under one minute. I would love feedback from heavy terminal users on what it would take for this to become your default memory of what you have been building breaking and shipping.
Are you familiar with the Unix `script` command?