Drydock brings Claude Code and Codex sessions into a native macOS app. Search across projects, resume work in real terminals, and get a briefing before diving back in. When you need another perspective, ask a second agent to review a snapshot or hand the work over with an editable context packet.
Screenshots show the actual 1.0 UI with sample session data.
Download the Apple Silicon .dmg from the latest release,
or install with Homebrew:
brew tap xd00099/tap
brew install --cask drydock --no-quarantineInstall the agent CLIs you want to use and sign in through their own tools. Claude Code is enabled by default; add Codex or another agent in Settings → Agents.
Drydock is not yet Apple Developer ID signed or notarized. If macOS reports the downloaded app as “damaged,” move it to Applications, then clear quarantine:
xattr -dr com.apple.quarantine /Applications/Drydock.appUpdates appear in the sidebar footer and preserve your open tabs after restart. Intel Macs can build from source.
Home brings together recent session summaries, milestones, and sessions waiting for your input. Start something new or jump back into a conversation without searching through terminal windows.
| Workflow | What you get |
|---|---|
| Find | Keyword and on-device semantic search, with proj:, starred:, and live: filters. |
| Resume | Real agent terminals, readable transcripts, Markdown export, and split panes. |
| Understand | Session details, task or plan status, token usage, and current Git file changes. |
| Organize | Projects, working folders, stars, and session names stored in Drydock. |
| Preview | HTML, SVG, and Markdown artifacts from Drydock-launched Claude sessions, with inline review. |
Open a session’s Briefing panel. Review with… sits with Uncommitted files, and Handoff… sits beneath Summary. This section lists staged, unstaged, and non-ignored new files across the session’s Git repository. With a clean checkout, it becomes Last commit files and shows the latest commit’s changes, matching the reviewer’s scope. It refreshes as files are edited or committed, including changes made outside that session.
The Summary section saves a short AI title and a few sentences about the session. Refresh updates both in one call through the source agent’s installed CLI. Manual session names always take precedence. Summaries show their update time and a New activity indicator when conversation content has changed. Codex summaries read the paginated history database as well as indexed rollout messages. They use actual user/agent text, excluding reasoning and tool payloads. Title-only metadata cannot generate a summary; without readable conversation content, refresh stays unavailable. Handoff… sits directly below the summary; Review with… stays with the Git file changes.
A new commit observed while Briefing is visible automatically refreshes its session summary after a short delay, combining rapid commits into one call. Opening a session, ordinary edits, and conversation messages do not trigger AI generation. Failed refreshes preserve the previous summary and can be retried. Automatic failures are not repeatedly retried for the same commit. There is no background sweep over other sessions. Refreshes use the CLI’s account usage.
- Review with… gives Claude Code or Codex a captured source snapshot, diff, and your acceptance criteria. Inspect findings against the captured files, copy selected findings, or prepare a fixes handoff. Changed source marks a review stale. Reviews suggest tests; they do not execute them or apply edits.
- Handoff… prepares an editable packet with the goal, progress, recent conversation, summary, and current diff. It refreshes the AI title and summary first, waiting for an existing refresh when needed. Failed refreshes show an error and retry; no stale handoff is prepared. Choose an agent and open a new interactive session. Source and destination stay linked.
Only enabled Claude Code and Codex profiles appear in these dialogs. Reviews use the installed CLIs and their account usage, save results locally, support cancellation, and time out after five minutes. Codex reviews exclude custom user config and use the standard provider/login. Interactive handoffs honor startup commands and share the original working files; pause other agents editing them.
Search and appearance
Search shows keyword matches immediately, then adds semantic matches when the local model is ready. The model unloads after two minutes of inactivity; document indexing also catches up at startup and twice daily.
Choose a dark or light palette in Settings → Appearance. UI controls and terminals follow the theme together; keyboard shortcuts are configurable too.
| Shortcut | Action |
|---|---|
| ⌘K | Search sessions |
| ⌘0 | Home |
| ⌘N | New session in any folder |
| ⌘T | New shell tab |
| ⌘⇧T | Switch between terminal and transcript |
| ⌘F | Find within the current pane |
| ⌘B / ⌘J | Toggle sidebar / session details |
| ⌘, | Settings, agents, themes, and shortcuts |
In the new-session dialog, arrows select a recent project and Tab fills its path for editing. Enter starts the session. Drag tabs to split the stage; ⇧⌘⏎ zooms a pane. Right-click sessions to organize, rename, or open their transcripts. Taking over a session running elsewhere asks before stopping it.
Project shows the current session's harness skills and MCP connections, including shared project tools and Drydock's native artifact and review tools. New and resumed Claude Code and Codex sessions receive the project's shared setup. Account sign-ins may still be required separately.
Claude Code and Codex have session discovery, resume, plain-JSONL transcript search, and session details. Enable pi, OpenCode, Copilot, Cursor, Droid, Gemini, or a custom command to launch them in terminal tabs.
Codex session details read successful plan updates and cumulative token usage when a readable rollout is available. File changes come from Git across the repository. Summaries and handoffs also read paginated Codex history directly; transcript display/search still require readable rollout files. Compressed rollouts are not yet imported. Checkpoint history remains Claude-specific.
Agent toggles and startup commands live in Settings → Agents. Advanced settings
live in ~/Library/Application Support/com.drydock.app/settings.json:
{
"claude_env": {},
"editor_cmd": "code"
}claude_env supplies environment overrides
to spawned processes; editor_cmd chooses the editor for changed files.
Drydock reads agent stores without changing their configuration. Explicit Delete permanently removes a transcript at your request. Its own index, organization data, and saved reviews live in its app-data directory. Search embeddings run on-device. Summary refreshes send bounded conversation excerpts through the source agent’s CLI. Reviews send captured code, and handoffs send the packet you preview, through the selected agent’s backend.
Other network activity includes update checks, the initial embedding-model download, configured MCP health checks, and allowed CDN resources in artifacts. Artifact previews restrict network destinations with a content policy; WebRTC is not covered by that policy. No telemetry is collected by Drydock.
Requires Rust stable and Node.js 20+:
git clone https://github.com/xd00099/drydock.git
cd drydock
npm ci
npm run tauri buildThe app lands in target/release/bundle/macos/Drydock.app. Local packaging may
require an updater signing key; for an unsigned local build, disable
bundle.createUpdaterArtifacts in a local Tauri config override.
For development, npm run tauri:dev opens Drydock Dev with a separate data
directory. Run npm test and cargo test --workspace before contributing.
Pushes to main run the tests and build the release binary to cache Rust
dependencies for later release tags. The first build after a dependency or
toolchain change may still need to compile dependencies. The shared cache also
includes the ONNX runtime used by text search; image-model codecs are excluded.



