Wake collects sessions from 15 coding agents into one searchable app

The developer iAmCorey has published Wake, a native desktop application that reads coding agent transcripts from 15 different tools and puts every session into one searchable interface. The repository has gained 731 stars and 48 forks on GitHub.

What it reads

Wake indexes sessions from Claude Code, Codex CLI, Cursor, OpenCode, Kiro, Gemini CLI, Grok Build, DeepSeek Harness, Hermes Agent, OpenClaw, Qoder, and several others. It groups the sessions by project directory, watches the transcript files for changes, and shows model information for most of the agents. Sessions on a remote host can be mirrored over SSH so that work done on a different machine appears in the same view.

Search and resume

The interface is built around a single search box, opened with the standard shortcut, that queries a SQLite full-text index of every transcript. The index handles code substrings and CJK characters, and search returns in under a millisecond on the author's benchmark of about 310 sessions. Each result shows the transcript with syntax highlighting for 30 or more languages, inline images, and collapsible tool calls. One click reopens the session in the original terminal at the original directory, which is the feature that gives the tool its name.

What else it does

Sessions can be starred, pinned, exported to Markdown, or moved to a local trash. A dashboard reports activity by day, by agent, and by project, with heatmaps and streaks. None of the data leaves the machine.

Where it runs

Wake is MIT-licensed and built in Rust with GPUI, the same UI framework Zed uses. The macOS build is a universal binary for both Apple Silicon and Intel. Linux support has been present since v0.2.5 and Windows since v0.2.7, both marked experimental. The project is at version 0.2 and publishing regular releases.

Anyone running more than one coding agent runs into the same problem: last week's useful conversation was in Codex, or maybe Claude Code, or maybe Cursor, and the transcript is either gone or hidden in a per-tool location that has to be reopened by the tool that wrote it. Wake reads the files directly and skips that step. For a team where different engineers standardise on different agents, it is also the first place a session done in one tool becomes searchable by someone who does not use that tool. The tradeoff is that Wake is at 0.2 and depends on the transcript formats of 15 tools that change on their own schedules, so the index can become out of date when an agent releases a new version. 731 stars in the first month says a real audience has the underlying problem.

Source

Primary source: iAmCorey/Wake on GitHub.