Skip to content

Latest commit

 

History

History
147 lines (117 loc) · 8.37 KB

getting-started.md

File metadata and controls

147 lines (117 loc) · 8.37 KB

Getting started

CodeConvoy runs one task across selected local Git repositories using one of your installed coding-agent CLIs. Start with a disposable repository or a small change you can inspect. CodeConvoy does not install agents, manage accounts or provide an additional security sandbox.

1. Install CodeConvoy and an agent

Follow the installation instructions for your OS. Check the exact release's requirements and SHA-256 checksums. The platform matrix distinguishes package checks from native desktop testing; Linux's glibc 2.39 target alone is not desktop certification.

Install Git and at least one supported CLI separately: Codex, GitHub Copilot CLI, OpenCode, Claude Code, or the experimental Cursor CLI or Antigravity CLI. Authenticate using that CLI's own instructions, then confirm it works in a terminal. Do not enter credentials in CodeConvoy tasks.

Launch CodeConvoy and select the agent in NEW CONVOY. Availability checks run in the background. If discovery fails, use Find CLI → Use and check, or enter the absolute executable path and choose Check CLI. Desktop and terminal PATHs can differ. On Windows use a native executable; shell wrappers are not supported. Available confirms the CLI contract, not authentication, account entitlement or access to a particular model.

Review the selected backend's options. Permissions are backend-specific; they are not interchangeable sandbox guarantees. See the contracts for Codex, Copilot, OpenCode, Claude Code and Cursor CLI and Antigravity CLI. Their sign-in and authenticated execution are separate from the help/version check. Claude Auto requires CLI/account support and can be disabled by policy; inspect its reported effective mode.

2. Register and group repositories

Enter an existing Git repository's root path, or select it with Browse…, then choose Add repository. Selecting a folder alone does not register it. Registration does not clone or modify the repository. Select its checkbox for the next convoy.

Use Manage groups… to save named collections. A repository can belong to more than one group but receives only one job per convoy. Selecting a group adds its available members; deselecting it removes those members even if another group overlaps. Check the final selection count.

Filter repositories… searches names and paths. Hidden selections remain selected. Select all / Select none and group checkboxes still affect their full sets, including hidden repositories. See groups, filtering and templates.

3. Prepare and start a convoy

Write the task once; every selected repository receives that task. Templates saves and loads task text only. Add files… supplies optional context according to the chosen backend's capabilities. Files must remain available and unchanged through launch; history retains references, not attachment contents.

Choose the execution mode:

  • Current working tree (default): the agent works directly in each registered checkout, including its existing edits.
  • Isolated worktree: a fresh detached checkout starts at the committed HEAD captured during review. Local staged, unstaged and untracked edits are excluded. Results stay in the retained checkout until an explicit result action. Configure new locations under Settings → Worktree location; existing attempts keep their original paths. See worktree settings.

Set This convoy and Global job limit. Both limits apply; conflicting repository jobs wait without consuming an execution slot. Choose Run Convoy, review the exact repository list, branch/HEAD and existing changes, and acknowledge dirty trees where required. Start convoy confirms execution. Git and context are checked again before each job starts; stale review data can fail a queued job.

The accepted run is an immutable snapshot. Editing the next draft does not change it. After launch, task text and repository selections clear for the next task; backend preferences and concurrency settings remain.

4. Inspect execution and Git changes

In RUNS / RESULTS, select a convoy and repository. Activity summarizes events, Raw output shows CLI details, and Task & settings shows the original request. Stop job or Stop Convoy cancels that work, waits for owned process cleanup and retains partial edits. Cancellation is not undo.

Open Review and Diff. Agent success, Git changes and validation are separate observations: a failed agent can leave useful edits, and a successful agent can leave mistakes. Direct diffs include pre-existing changes; isolated diffs use the captured base. Use Refresh review / Refresh diff after external edits. Review the actual Git changes before committing through your normal Git tools.

For an isolated result, Apply result can copy reviewed changes to the original checkout only when ownership is verified and the destination is clean at the original base. It leaves unstaged changes, without committing or pushing. Unsupported or uncertain changes stay blocked with diagnostics. Discard result deletes a verified retained checkout after confirmation. Applied copies remain until Clean up retained copy is explicitly confirmed. These result actions are distinct from conservative Storage cleanup; Direct mode has no Apply/Discard. See result safety and cleanup rules.

5. Run manual validation

In Repositories, choose Configure validation…, select a preset or enter an executable and a JSON argument array, then Save command. For example, Rust Tests configures cargo with ["test"]. Assign Validation Preset… can configure several repositories/groups; it preserves existing commands unless you explicitly confirm overwrite. Configuration never executes a command.

Select a finished repository job and choose Run Validation in Review or the Validation tab. Check the displayed directory and command: Direct uses the recorded source, isolated uses the retained worktree. Applied or unavailable results cannot validate. Inspect the output and use Cancel Validation if needed. Commands can change files; a pass records that attempt, not a guarantee about later edits. See validation and platform requirements.

6. Revisit history and manage storage

Completed convoys appear under HISTORY. Search/filter does not change the selected result. Reopening the app never resumes an agent; interrupted work is labelled accordingly, and retained worktrees need fresh verification. Reuse convoy restores a draft for a new review, not a running session.

Copy Summary / Copy Repository Result copy bounded Markdown reports. Export… / Export History… save Markdown or JSON reports of recorded metadata. Review them before sharing; names, task text and validation arguments can be sensitive. Reports omit logs, diffs and ownership records and are not backups. Output logs and validation output are available only during the current session.

History is unlimited by default. Settings → History Retention… configures optional limits; History cleanup → Clean Up History… previews metadata removal. Protected work and retained ownership records stay protected. Removing history never deletes a checkout. See history and export.

Settings → Storage… measures known history/worktree storage. Manage Worktrees offers separately verified, confirmed cleanup of eligible entries. Unknown, modified, active and unresolved entries remain protected. Ordinary Storage cleanup is unavailable on Windows; measurement and inspection still work. See Storage Overview and safe cleanup.

Desktop controls

Use Settings → Keyboard Shortcuts… for the in-app reference, or the shortcut table. Settings → Desktop notifications controls completion alerts. Tray support is optional and off by default; close-to-tray needs a working tray. Explicit Quit confirms active work and waits for cleanup. See desktop behavior and notification limits, including Wayland focus restrictions.