@tryjourney/cli
v0.5.0
Published
CLI for recording and retrieving engineering decisions with Journey
Maintainers
Readme
Journey CLI
Journey records important engineering decisions, problems, and resolutions as structured events that coding agents can retrieve through MCP.
Install
npm install --global @tryjourney/cliUse
Authenticate with Journey, then configure an individual project for Codex or Claude Code:
journey login
cd your-project
journey initjourney init configures a local MCP gateway and adds concise Journey
instructions without replacing unrelated project configuration.
Commands
journey init Configure Journey for the current project
journey login Connect a hosted Journey account
journey logout Clear the hosted session
journey status Show the current hosted account
journey scan Scan locally (`--deep` and `--review` are supported)
journey sync Reconcile local and hosted architecture (`--deep`, `--force`)
journey architecture explain <id> Show provenance and evidence
journey architecture feedback <id> --confirm|--reject|--clear Manage a local correction
journey architecture feedback --export Export local corrections as JSON
journey doctor Check binding, authentication, API, and revision health
journey rebind Bind the repository to a different existing project
journey delete Delete the bound project and local Journey setup
journey mcp Run the local authenticated MCP gatewayjourney publish is reserved for a future Journey feature.
Delete a project
Deletion is available from CLI 0.5.0:
journey delete --dry-run
journey deleteThe command displays the bound project and requires typing its name. It deletes every session and all hosted history for that project, then removes this repository's binding, scans, pending uploads, feedback, Journey MCP entries, and marked instructions. Source files, other agent settings, other projects, account credentials, and the installed CLI remain available. Customized Journey entries or malformed configuration must be resolved before combined deletion.
journey delete --local-only # Remove local setup; keep hosted data
journey delete --remote-only # Delete hosted data; keep paused local setup
journey delete --yes # Explicit confirmation for automationStop existing Journey MCP sessions before deleting, especially sessions started
with an older CLI version. Remote-only deletion disables further synchronization.
If hosted deletion fails, local setup is preserved. If the remote response is lost,
retry journey delete; if hosted deletion succeeds but local cleanup fails, run
journey delete --local-only to finish after resolving the reported paths.
The web project's Delete project action removes hosted data only. Run
journey delete --local-only in each clone afterward. Other devices stop remote
activity when they discover that the project is no longer accessible. Creating a
replacement requires an explicit journey init after cleanup. Deletion cannot be
undone through the app; infrastructure backups follow their configured retention.
Requirements
- Node.js 20 or later
- A Journey account at app.tryjourney.dev
Links
Project architecture observation
journey init binds the current Git root to an owner-scoped project, writes the agent configuration, performs an initial metadata scan, and prints the project URL. journey mcp continues observation while the coding agent is connected. Run journey status to inspect the binding, last scan, last sync, queue depth, and observer errors.
Set JOURNEY_LOCAL_ONLY=1 when initializing to keep snapshots in the platform user data directory. Add paths or glob patterns to .journeyignore to exclude them from scanning. The observer uploads derived architecture metadata and relative evidence paths; it does not upload source contents, .env values, absolute paths, or raw conversations.
The default scan is deterministic. journey scan --deep or journey sync --deep
explicitly opts into semantic analysis through the locally installed Codex CLI.
Codex runs with a read-only sandbox and schema-constrained output; only proposals
that pass Journey's evidence verifier enter the canonical graph. Codex may send
files it chooses to inspect to the configured OpenAI service. Deep analysis never
runs in the periodic observer and falls back to the deterministic graph on failure.
