Skip to content

Roadmap

Where BenchBar is and where it goes. Versions below 1.0 may change the JSON API and the runner format; 1.0 freezes both. Dates are not promised. Ideas are welcome as issues.

Released

0.1, the CLI. Benches under launchd with one agent each, a crash guard (three restarts in ten minutes, then pause and notify), resume after reboot only when the bench was running, doctor and repair for the common breakages (missing env, node modules or built assets, crash looping agents, a stray Redis, MariaDB bound to the network, CleanMyMac), idempotent installs, the lean Procfile and the bench* helpers.

0.2, the menu bar app. The rename to BenchBar with the frappe-mac alias kept and old agents migrated, a versioned JSON API, the animated runner that sleeps, runs, speeds up with load and stumbles on crashes, start, stop, restart, site, logs and a read only doctor from the popover, crash notifications, launch at login, Reduce Motion, custom runners.

0.3, easy install. The one line installer, the manual install steps automated (MariaDB root password in the Keychain, the secure installation in SQL, the utf8mb4 drop-in, the pinned and checksummed wkhtmltopdf with Rosetta offered, the /etc/hosts line inside markers, one sudo prompt per run), benchbar adopt for existing benches, benchbar report for redacted bug reports, CI on macOS and Linux, unsigned releases built as drafts from a tag.

0.4, Frappe v16 and more than one bench. The v16 profile tested end to end on a real Mac, one MariaDB server shared by every bench, several benches side by side with their own port blocks, settings and processes, sites (site add, site default, site hosts), the scheduler opt in, doctor checks from the community threads (Full Disk Access, the toolchain as the bench sees it, honcho without pkg_resources, stale processes), and the app with a bench list, the worst state in the menu bar and the sites of each bench.

0.5, the app does more, apps and team profiles. The BenchBar window with a page per bench (sites, apps, doctor and Repair with its plan first), a log window, benchbar repair --json, benchbar mcp for coding agents, app installs and updates from the registry or any GitHub repo (private ones too), team profiles kept outside BenchBar, the benchbar.toml team lockfile, benchbar pull for a production copy with the encryption key carried over and email and the scheduler off, a new app icon, and the macOS 27 look.

0.5.5, the project skin. The documentation site at benchbar.akashmishra.com with a page per command, a short README, the community files (contributing with an AI policy, code of conduct, security policy, issue forms), and in the app an About pane with an update check, a Help menu and Report a Bug; benchbar docs and doctor links into the docs.

0.5.6, cleanup tools. Doctor warns about Mole until the bench is in its whitelist, finds CleanMyMac installed from Setapp, and no longer flags the installer’s PATH block.

0.5.7, find your benches. Scan Folder in the menu bar and Find Benches in the app find existing benches in a folder, remember the ones you pick, and set up their service after showing the plan.

0.5.8, quick wins. benchbar:// links for Raycast and Shortcuts, CPU and memory charts per bench, site backup, backups and drop with a backup first, the lockfile badge, Open in VS Code or Cursor, benchbar console and db, doctor --fix-hints, and port blocks planned across benches with a check before every start.

0.6, sharing profiles between teams. benchbar profile export writes a portable copy (SSH host aliases resolved, each app on its repo’s default branch after a review, public, private and personal repos marked, the apps each one requires listed), profile import takes a file or an https URL and checks access to every repo first, profile subscribe follows a team’s config repo with an outdated warning instead of silent pulls. Doctor warns when an app your work depends on falls behind, never about the app you are working on. The app checks for updates once a day and updates in one click, the CLI with benchbar self-update.

0.6.1, cost at rest, signed. Under 0.1 percent of a core at rest (from about 6): status in 5 programs, a runner heartbeat, an app that polls on events. The first release signed with a Developer ID, notarized and stapled, with Sparkle updates.

0.7.0, Homebrew. brew install askysh/tap/benchbar askysh/tap/benchbar-app installs the CLI and the app from askysh/homebrew-tap, and every release updates the tap. The CLI keeps its state in ~/.local/state/benchbar, records paths that survive brew upgrade, and updates with brew upgrade; benchbar repair moves a one line install over, and doctor warns about a second CLI or app.

0.7.1, one CLI. BenchBar.app carries the CLI of its own version, and Homebrew’s and the one line installer’s benchbar hand off to it, so every benchbar on the Mac is the app’s and one update covers both. Update Now goes through Sparkle without Terminal; benchbar where names the copy that handed off.

0.7.2, a calmer window. One main action per page, the rest in a ⋯ menu, and one layout for every sheet (Return confirms, Esc cancels).

0.7.3, the fix pass. Nine batches of review findings, checked on a real Mac before the release. benchbar stops only processes it can prove are the bench’s, and the runner pauses with port_conflict instead of starting into a taken port. v15-lts moves to node@22 before Homebrew disables node@20 on 2026-10-28, with a Formula lifecycle check and a release gate. A fresh Homebrew MariaDB with a socket only root no longer stops the install. The env is never rebuilt on a guessed profile or while the bench runs. Long commands run in their own process group, and every run that changes a bench takes that bench’s lock. Writes fail closed, step results say what really happened, secrets stay out of child processes and logs, and paths are checked before they reach root. The MCP server runs calls on threads, checks every argument against its schema and labels third party text. The installer’s state moves in even when the app made the folder first. The app names a port conflict, and doctor --json says where the profile came from.

0.8.0, a bench without Terminal, Linux, a Windows preview and agent contracts. A first run wizard in the app: check the Mac (doctor’s new prerequisites), create a bench from a built in or team profile, or find and adopt the ones you have, with every step streamed (install --json, adopt --json) and the two privileged steps in macOS’s own password dialog instead of sudo in Terminal. Benches are the home of the BenchBar window, with an Add Bench menu, Settings and About in the sidebar footer (Settings and About are windows of their own), and the app explains itself where you are: a four step walkthrough, why a bench is stopped and what to do next, empty lists that say how to fill them. The CLI runs on Linux (Ubuntu 24.04, v15-lts) with each bench under systemd --user. On Windows, a preview: benchbar.exe forwards the CLI from PowerShell to WSL, BenchBar Tray shows the benches in the notification area, and a zip per CPU with a one line installer is attached to each release (unsigned until the signing application is approved). Agents get a contract: benchbar COMMAND --help and commands --json, one JSON stream for every command that changes things ending in a done line (exit, changed, warnings, next_actions, state_after), --for-agent, a non interactive mode that names every missing input, doctor --ack and --quick, and logs with a time per line. An app update that stops half way is finished by the next one.

0.8.1, Windows beta and Open as. The Windows tray takes a new PC from no WSL to a running bench and manages benches from its own window, with Report a Bug everywhere. benchbar open --as USER and Open as… in the app sign a development site in as any of its users, off until a person turns it on per site (benchbar site open-as on). Settings and About are panes of the one BenchBar window again, a live QA pass fixed the popover in dark mode and the Find User flow, and the menu bar has new runners. Behind the scenes: work lands on next-release and a release tags itself when it reaches main, and the mocked test suite runs about a quarter faster.

0.8.2, Labs. Experiments ship behind one switch that the CLI owns (benchbar labs, Settings > Labs in the app). The first two: rooms, which share a hardened copy of a dev site with people on the same Wi-Fi behind a room code, with lan-guard keeping the bench ports off the LAN, and Quick Switch (⌃⌥U), which opens a site as another user from any app. Open as has keyboard shortcuts in the popover.

0.8.3, Windows fix pass. A full review of the Windows side and a live test on a Windows 11 PC, from an empty PC to a running bench: the installer checks Smart App Control before it writes anything, removes the internet mark from what it installs and can uninstall; benchbar.exe puts deadlines on WSL calls and explains WSL errors; the tray’s setup wizard recovers from more states; and install inside WSL checks systemd, ports, the network and the bench folder before it asks for sudo. The Windows programs are still unsigned.

Launch. The post on discuss.frappe.io went up on 2026-10-04: BenchBar: local Frappe benches on macOS with a menu bar app.

Later

The app’s runner and CLI link. The app’s own runner stops a command’s whole process group, as the CLI does since 0.7.3, and the app makes its CLI link again when a poll finds it missing.

Management in its own target. Team profiles, bench discovery and port setup move out of the menu bar app into a separate management target that runs only while it is open. The reason is scope: the menu bar app should start, stop and show benches. Not energy: 0.6.1 brought the app to under 0.1 percent of a core at rest with these in it.

0.8.x, the app for sites. Restore from the app (backup and drop shipped in 0.5.8), and pull and lock apply in the app.

0.9, agents. What the agent contracts round of 0.8 left out on purpose, so nobody reopens it in a pull request: operation handles for long jobs (an op_id, benchbar op status, MCP progress notifications fed by the stream’s progress lines; the MCP 2025-11-25 tasks primitive is the eventual home), benchbar run --site X -- migrate|run-tests|... with an allow list, site snapshots and rollback, scratch benches pinned to commits with a TTL and bench gc, leases per bench or site, benchbar history --json, bench remove, bench forget, site move --to-bench, site hosts --prune, db orphans, and uninstall-service scoped to one bench. Two measurements come first as well: a comparative benchmark of native versus container benches (idle memory, build time, file watch latency, startup) before the README keeps its performance wording, and a four week pilot with about ten developers across three teams.

Windows past the preview. Signed binaries once the SignPath Foundation application is approved. The preview of 0.8.0 runs the benches inside WSL with the Linux platform layer (uv Python, fnm Node, MariaDB and Redis from apt, each bench under systemd --user), a benchbar.exe shim and a tray app.

1.0. A stable JSON API and runner format, an official Homebrew cask, full doctor coverage for v15 and v16. Vouch (github.com/mitchellh/vouch) when drive-by PRs appear. Not before.

Ideas

Not scheduled, kept because they came up more than once.

  • Shortcuts actions, a Raycast extension, desktop widgets (the URL scheme shipped in 0.5.8).
  • A local mail catcher for development email.
  • Run one scheduler event now from the app.
  • Worker restart when Python files change, opt in.
  • More speed sources for the runner: job queue depth, requests per second.
  • A runner gallery in the docs.
  • Log rotation on a size limit without a manual repair.
  • More failure path tests for bench creation and app installation.
  • benchbar wipe, the uninstall recipe behind an explicit confirmation, never touching MariaDB data without a backup.
  • Intel Mac verification of the CLI.

Not planned

  • Docker or VMs. A bench runs natively: Python, Node, MariaDB and Redis from Homebrew, the processes under launchd. File watching, bench build and debugging are faster than through a VM, there is no Docker Desktop license or memory overhead, and the setup matches what most Frappe developers run on Linux.
  • The Mac App Store. App Store apps run in the App Sandbox, and a sandboxed app cannot run the CLI, start launchd agents or read a bench in your home folder. BenchBar ships as a signed, notarized download and a Homebrew cask instead.
  • Production deployment. BenchBar is for development benches.