Wake gathers coding agent sessions on your own computer

What this is about

Using several coding agents quickly spreads work across many directories, databases, and conversation histories. Wake gathers these locally stored sessions in a desktop app. Users can search by agent, project, and time, read full transcripts, and resume work from the appropriate project directory.

The open source project by Corey Chiu first appeared in August 2026. Version 0.9.3 was released on October 13, 2026. Wake is available under the MIT License. macOS 14 or later is the primary supported platform, while Linux and Windows builds are described by the project as experimental.

What Wake actually does

Wake reads history files and local databases from different tools. These include Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Kiro, OpenClaw, and ZCode. The app builds its own local full text index. Titles, conversation content, tool calls, and model details can then be searched together.

Wake can open a found session in a terminal or an appropriate desktop app. Stars, pinned items, and the search index are stored in Wake's own database. According to the project, source files are read only. If a user explicitly deletes a file based session, Wake moves it to the Trash.

The app also ships with a local MCP server and a command line interface. Compatible agents can use them to search and read previous sessions. Since version 0.9, new Claude Code and Codex sessions can receive context from earlier project work through a plugin. Version 0.9.2 also groups sessions from many subagents under their parent session.

Why it matters

Coding agents do not store histories in one shared format. A project may start in Claude Code, continue in Codex, and later move to Cursor. Without a central search tool, a person must remember which tool was used and where its history lives. Wake targets this organizational gap.

The practical value is strongest for developers who test several agents, work on many projects in parallel, or need to recover an earlier decision and its reasoning. Local processing is also an important distinction: the product page says there is no account, subscription, or telemetry. Only a manually triggered update check makes a background request. Remote machines are connected through the user's own SSH connection.

For the developer's machine, the project reports roughly 1,100 sessions and about 1.8 gigabytes of source data. A complete index reportedly took around five seconds there. This is a vendor statement, not an independent benchmark; results may differ on other computers and with other data formats.

In plain language

Wake is like an index card cabinet for workshop jobs. The actual work stays in the drawers of the individual tools. Wake creates a shared catalog so you can find the right job and continue at the appropriate workbench.

A practical example

A developer works on four projects during one week and starts 30 sessions across Claude Code, Codex, and Cursor. On Friday, she needs the earlier rate limiting decision but cannot remember the agent or the day. She searches Wake for “rate limiter,” opens the result containing the original tool call, and resumes the session from the correct project directory.

Before enabling the MCP plugin, she checks several random results and the displayed project paths. She then gives the new agent read only access to earlier sessions. This provides useful context without automatically treating every old assumption as still correct.

Scope and limits

First, Wake can only read histories stored locally in a supported format. Windsurf and Trae are unsupported because their data is encrypted, and cloud only histories from other providers may also be unavailable.

Second, version 0.9.3 is still before a stable 1.0 release. Linux and Windows are explicitly experimental. Teams should test it on a copy of their working environment and should not treat Wake as their only archive.

Third, session histories often contain source code, file paths, prompts, or internal information. Local processing reduces transfers to third parties, but it does not replace access controls, disk encryption, and clear deletion rules. Connecting remote computers through SSH and exposing old transcripts to new agents should be deliberate choices.

SEO and GEO keywords

Wake, coding agents, Claude Code sessions, Codex sessions, Cursor history, local full text search, MCP server, developer tool, open source, session manager, agent memory