Atuin Threads syncs your AI agent sessions across machines

Atuin has launched Threads, a feature that pulls the sessions of AI coding agents into one searchable, synced history you can pick up on any machine. For developers who now spend much of the day prompting agents rather than typing shell commands, it targets a gap that shell history never filled: a record of what an agent did, and why.

According to Ellie Huxtable, writing on the Atuin blog in a post dated 27 October, the tool grew out of her own frustration with how her work had changed.

What is Atuin Threads?

Atuin started as a shell-history tool: a searchable, permanent alternative to the .bash_history file that, in Huxtable's telling, was always going to discard her older commands. Threads extends the same idea to a newer kind of work record — the sessions produced by AI coding agents.

The problem it addresses is fragmentation. Huxtable writes that more than half the commands running on her machines are no longer typed by her but generated through coding agents, and that those sessions are scattered across different tools and machines — a laptop, a dev server, temporary cloud instances, a cheap bare-metal box — hard to search and easily forgotten. Shell history, she argues, records commands but not the reasoning or intent behind them. Agent sessions capture all of it: the commands, their output, the intent, and the false starts, in plain English.

Threads makes that record portable. Starting in Atuin 18.24, you can search agent sessions across machines, resume them anywhere, or fork a session into a different agent harness — begin a task in Claude Code on a dev server, then finish it in Codex on a laptop, with context intact. Atuin published a short clip demonstrating that handoff. Watch the resume demo.

How does Threads sync and protect your agent sessions?

Rather than build a new sync system, Atuin reused the encrypted sync engine it had already developed for shell history. Huxtable describes that engine as generic: it can sync anything that can be modelled as an append-only log, and, as she puts it, most AI sessions turn out to be exactly that.

Privacy works the same way as with shell history. Synced Threads are end-to-end encrypted, so Atuin says it cannot read your sessions. You choose where the data lives: Atuin's hosted sync server, your own server, or nothing leaves your machine at all. Capturing agent sessions is opt-in, so the feature does not record anything until you enable it.

Which coding agents does Atuin Threads support?

At launch, Threads supports five agents — Claude Code, Codex, Pi, Omp and OpenCode — with more promised. Because the underlying format is a generic append-only log, adding agents is presented as an extension of existing work rather than a rewrite.

Threads also ships a Model Context Protocol (MCP) server. That lets agents search and reference your previous sessions on their own: finding earlier work, understanding why a past decision was made, and building on prior investigations without a person digging up the context. The MCP is available through atuin mcp, and can be installed for a specific agent with atuin mcp install claude/codex/pi/opencode.

What can you do with a synced agent history?

Huxtable offers three examples from her own use.

The first is recovering work that never shipped. After many attempts at redesigning an onboarding flow, she says, an agent newly asked to work on onboarding can find all the earlier attempts, why each was abandoned, and what to do better — including code that never reached the main branch, plus the original prompts. A later agent can infer the goal without a hand-written spec.

The second is faster incident response. After rolling security keys as a precaution, one new object-store key lacked the permissions needed for avatar uploads, and the issue went unnoticed at first. Her daily Sentry-triage agent traced the failure to a command run earlier that day; prompting another agent with a short note about "tigris avatar issues" led it to the earlier investigation, the cause and a fix faster than starting cold.

The third is forking a session across machines. While testing Threads, one machine underperformed, and the investigating agent suspected slow disk I/O but needed faster hardware to confirm. Huxtable forked the session onto a MacBook, where it resumed exactly where it left off; the faster SSD showed slow disk was only part of the story, and the original machine's thread later pulled those benchmark results back to build a fix — no copy-paste or manual coordination required.

How do you install and start using Threads?

Threads ships in Atuin 18.24 and later. Atuin can be installed or updated with its setup script:

curl --proto '=https' --tlsv1.2 -LsSf https://setup.atuin.sh | sh

Running atuin resume walks you through setup and opens the Threads interface, and atuin mcp install wires the MCP server into your agent. Huxtable also flags a team feature in progress — opt-in session sharing, so a thread can be handed off "just like a Git branch" — and invites teams who use agents daily to get in touch.

How does this relate to Grafy?

Threads is a terminal tool for command-line coding agents, so it does not map one-to-one onto Grafy. But the principle underneath it — that the history and reasoning behind AI-assisted work should be searchable, portable and yours to keep — is one any developer using AI coding tools will recognize.

If you write code with AI, Grafy Code is a browser-based AI coding studio where you describe a task and it generates and runs code — a different setting from a local terminal, but the same broad category of tool. For a closer look at how it works, see An AI coding studio that installs packages and runs your code.

The memory angle is worth noting too. Threads earns its keep because agents can recall the decisions and dead ends behind past work; Grafy's take on keeping that context available over time is Project memory: an AI that remembers your decisions.

Frequently asked questions

Can you resume an Atuin Threads session on a different machine?

Yes — that is the point of Threads. According to Atuin, you can start a task in one agent on one machine and resume or fork it elsewhere with the context intact, because sessions sync through Atuin's engine. The published demo shows a task begun in Claude Code being picked up in Codex on a laptop, illustrating the cross-machine handoff the feature is built around.

Is agent-session capture turned on by default in Threads?

No. Atuin states that capturing agent sessions is opt-in, so nothing is recorded until you enable it. When you do sync, the data is end-to-end encrypted, and Atuin says it cannot read your sessions. You can also choose to keep everything local rather than syncing to any server, which the post lists as one of three hosting options.

Does Atuin Threads replace your shell history?

No. Huxtable frames Threads as an extension of the idea that started Atuin, not a replacement for shell history. She argues shell history remains useful — arguably more so than ever — while Threads adds the reasoning and intent that shell history never captured. Both, in her framing, are meant to stay searchable, accessible and yours to keep.

Source: Atuin — Atuin Threads: all your agent sessions on every machine.

---

Source: Hacker News — front page — Atuin Threads: All your agent sessions on every machine