Atuin Threads: All your agent sessions on every machine

Atuin, a tool that keeps a searchable history of the commands you type in your terminal and syncs it across your machines, now does the same for sessions with AI coding agents (Atuin blog). The feature, called Threads, arrives in Atuin 18.24. It captures sessions from Claude Code, Codex, Pi, Omp and OpenCode, makes them searchable across machines, and lets you resume a session on another machine or fork it into a different agent, for example starting a task in Claude Code on a dev server and finishing it in Codex on a laptop.

The post's argument is that an agent session is a richer record of work than shell history. A history file shows which command ran; a session also holds the command's output, what the user was trying to do and the dead ends along the way, in plain English. Those sessions usually sit in files scattered across tools and machines and are soon forgotten. Threads reuses the sync engine Atuin built for shell history, which can sync anything stored as an append-only log, a record that only ever grows by adding entries to the end, and most agent sessions take exactly that form. Synced sessions are end-to-end encrypted, so Atuin cannot read them, and users can use Atuin's hosted server, run their own or keep everything local. Capturing sessions is opt-in.

Threads also ships an MCP server, a standard interface through which an agent can call outside tools, so agents can search earlier sessions themselves. The examples come from the author's own work. An agent asked to work on onboarding found earlier abandoned attempts and the reasons they were dropped. After an avatar-upload failure caused by a storage key without the right permissions, a second agent found the investigation the author's daily triage agent had already run, which traced the failure to a command run earlier that day, and produced the fix much faster than starting from scratch, by the author's account. These are anecdotes, not measurements, but they show the intended use: past agent work becomes context the next agent can look up instead of something a person has to explain again. Sharing sessions across a team is still being built.