Atuin v18.13 adds AI shell commands, faster search daemon, and PTY proxy

✍️ OpenClawRadar📅 Published: March 21, 2026🔗 Source
Atuin v18.13 adds AI shell commands, faster search daemon, and PTY proxy
Ad

Search improvements with daemon

The daemon, previously experimental, now maintains a hot, in-memory search index powered by a modified version of nucleo, which implements the same algorithm as fzf. Configurable multipliers for frequency, recency, and frecency scoring are available. Enable with:

search_mode = "daemon-fuzzy"
[daemon]
enabled = true
autostart = true

With autostart enabled, Atuin manages the daemon lifecycle. The daemon also enables syncing records without shell interaction, keeping remote machines updated.

Atuin AI for shell commands

Press ? on an empty prompt, describe what you want, then press enter to execute or tab to edit. Press f to follow up with adjustments or questions. Enable with:

[ai]
enabled = true

Accuracy comes from frontier models combined with a dataset powered by man pages and command outputs. Safety checks flag dangerous commands (like wiping data or restarting deployments) requiring double enter to execute. Data privacy is minimal by default—only OS and shell info—with optional permissions for directory path, contents, git status, etc.

Ad

PTY proxy (hex)

Hex enables Atuin popups to render over previous terminal output without clearing it, unlike previous fullscreen or inline modes. It's a lightweight PTY proxy that behaves like minimal tmux but doesn't affect scrollback or break terminal features. Enable one-off with atuin hex or permanently with:

eval "$(atuin hex init)"

Future uses include searching/syncing command outputs, supporting shells without hooks, and readline-esque features.

Other updates

New authentication for hosted sync service (Atuin Hub) adds Google/GitHub login and email recovery. Existing credentials still work. Hub isn't open source yet due to rapid changes. Self-hosting remains unchanged.

📖 Read the full source: HN AI Agents

Ad

👀 See Also

Claude Code's Illusion of Finished Work: Why Reviewing the Agent's Path Matters More Than the Diff
Tools

Claude Code's Illusion of Finished Work: Why Reviewing the Agent's Path Matters More Than the Diff

Claude Code can produce a clean diff, passing tests, and a good summary—yet still miss real behavior, security concerns, or architecture constraints. The author argues that reviewing the chain of actions (plans, files read, commands run, test output) is now essential, not just the final diff.

OpenClawRadar
Gemma4 26B-A4B Delivers Fast Local Performance with Web Search and Image Support
Tools

Gemma4 26B-A4B Delivers Fast Local Performance with Web Search and Image Support

The gemma-4-26B-A4B model achieves approximately 145 tokens per second on an RTX 4090 and includes web search MCP and image support for chat applications. A blog post details setup and cross-platform usage on Mac and iPhone.

OpenClawRadar
Dual-model architecture reduces token consumption by half for long conversations
Tools

Dual-model architecture reduces token consumption by half for long conversations

A developer built a dual-model system where a small 'subconscious' model compresses conversation history in the background, allowing the main model to work with a curated ~35K context instead of 120K tokens of raw history. This architecture cuts token consumption roughly in half for sustained project work.

OpenClawRadar
Vibeyard Adds Kanban Board for Managing Multiple Claude Code Sessions
Tools

Vibeyard Adds Kanban Board for Managing Multiple Claude Code Sessions

An open-source IDE called Vibeyard now includes a Kanban board that lets you spin up Claude Code agent sessions directly from cards. Cards auto-move to Done when the agent finishes.

OpenClawRadar