Wake has garnered 684 stars on GitHub just weeks after its public launch. This desktop application, written in Rust, solves a specific problem: your sessions with Claude Code, Codex, and other code agents are often trapped in scattered, hidden directories across your system, making unified access impossible. Wake reads these histories—in read-only mode—allowing you to search, visualize, and resume any conversation from a single, native interface.
Where Do Your Code Agent Conversations Get Stored?
Each agent tool stores its history in different proprietary paths. Claude Code saves its sessions in ~/.claude/projects/, Codex CLI uses ~/.codex/sessions along with an SQLite database, and Qoder CLI writes to ~/.qoder/projects/. This fragmentation forces developers to remember which agent they used for each task and navigate hidden directories to retrieve lost context.
This dispersion becomes more pronounced when switching between multiple agents on the same project. Wake identifies up to fourteen distinct local data sources, from Grok Build (~/.grok/sessions/) to DeepSeek Harness (~/.dsh/sessions/), including OpenCode, Kiro, and Pi. The application indexes these files—primarily JSONL and SQLite—without altering them, creating a centralized catalog that reflects your actual activity.
| Agent | Typical Local Path | Model Visible | Source Visible |
|---|---|---|---|
| Claude Code | ~/.claude/projects/**/*.jsonl | Yes | No |
| Codex CLI | ~/.codex/sessions + state_5.sqlite | Yes | Yes |
| OpenCode | ~/.local/share/opencode/opencode.db | Yes | No |
| Cursor (CLI) | ~/.cursor/projects/**/agent-transcripts | No | No |
| DeepSeek Harness | ~/.dsh/sessions/**/session.jsonl[.zstd] | Yes | No |
How Does Wake's Search Index Work?
Wake builds a full-text index using SQLite FTS5 with trigram support, enabling instant searches for code substrings like useEffect(. For the author's team, indexing approximately 310 sessions totaling 800 MB of JSONL files takes about five seconds on the first run; subsequent launches are immediate due to incremental scanning based on mtime.
The integrated search engine—accessible via ⌘K or Ctrl+K—handles Chinese, Japanese, and Korean (CJK) text in addition to source code. When a result is selected, the application jumps directly to the specific message within the full transcript, rendering user and assistant bubbles with syntax highlighting via tree-sitter for over thirty programming languages.
What Does It Mean to Resume a Session Without Losing Context?
Wake doesn't just display history; it reopens the exact session in your preferred terminal within the original working directory. By clicking the resume button, the application launches the appropriate command—claude --resume, codex resume, or another agent-specific command—inside iTerm or Terminal on macOS, or in the native emulator on Linux and Windows.
This feature eliminates the friction of manually reconstructing the state of an interrupted conversation. As the repository itself explains, "Wake reads them all, read-only, and gives you one fast window into it." Developers can pick up the agent's thought process without copying and pasting context between windows.
Is It Safe to Index Conversations Containing Private Code?
Wake adopts a strict local privacy stance. Agent data directories are opened exclusively in read-only mode; the application never writes to original files or accesses credentials like auth.json. The only network connection occurs when the user explicitly checks for updates against GitHub Releases metadata.
Wake's own database resides at ~/Library/Application Support/wake/wake.db on macOS, ~/.local/share/wake on Linux, or %LOCALAPPDATA%\wake on Windows. It can be rebuilt from scratch at any time without permanent data loss, as favorited or pinned items are stored in separate tables that survive reindexing.
What Operating Systems Does Wake Currently Support?
The application prioritizes macOS with a universal binary compatible with Apple Silicon and Intel architectures. Since version 0.2.5, it includes experimental support for Linux (arm64 and x86_64) via .deb packages and tar.gz archives with a user-level installer. Version 0.2.7 adds experimental support for Windows.
Building from source code requires the Rust toolchain and relevant system dependencies: libasound2-dev, libfontconfig1-dev, libwayland-dev, and libzstd-dev on Debian/Ubuntu. The scripts/make-app.sh script generates the ad-hoc signed bundle for macOS, while CI workflows prepare artifacts for Debian-derived distributions.
How to Integrate Wake into Your Development Workflow
Centralizing code agent session management is the logical next step after adopting tools like Claude Code or Codex CLI. If you already use these agents daily, Wake eliminates the cognitive load of remembering where each conversation left off. You can delve into specific Claude Code usage in our guide on how to use Claude Code for assisted programming or explore alternatives like Codex in the complete Codex CLI tutorial.
The application allows exporting any session to Markdown for external documentation, deleting obsolete conversations—sending them to the system trash with tombstone logging—and visualizing usage statistics through GitHub-style heatmaps that reveal work patterns by hour, day of the week, or month.
Frequently Asked Questions
Does Wake modify original Claude Code or Codex files?
No. Wake accesses the ~/.claude, ~/.codex directories, and other agent paths in read-only mode. It never writes to their databases or JSONL files, only reading and indexing content into its own local database.
Does Wake work with Cursor, Windsurf, or the IDE versions of these agents? No. Cursor IDE, Windsurf, and Trae encrypt their local data, preventing third-party access. Wake does support CLI versions of some agents when they store transcripts in plain text, but it cannot access cloud-hosted sessions like those from Warp or Factory (Droid).
Can I use Wake on Windows or Linux? Yes, though it's experimental. Versions 0.2.5 and 0.2.7 introduced preliminary support for Linux and Windows, respectively. macOS remains the primary and best-supported platform, available as a universal binary for Intel and Apple Silicon.
How does Wake find text within long conversations? It uses SQLite FTS5 with trigram indexes. This allows searching for exact code snippets—such as specific function calls—or text in Asian languages (CJK). Results appear in less than a millisecond, according to tests in the official repository.
Do I need to pay for Wake or subscribe to a service? Wake is free software under the MIT license. It requires no subscription or user account. The only optional network connection is manual update verification against GitHub Releases; all indexing and search functionality works 100% offline.