Roadmap

In rough order of how settled it is. Anything here can be reordered by what people actually hit.

• More harnessesThe registry was built so a new agent CLI costs a descriptor and an adapter instead of a branch through every feature. The candidate study exists; spending that abstraction is the work.
• macOSImplemented behind the same platform seams as Linux and typechecked on every gate run, but never executed. It becomes a supported platform when somebody runs the suite on real hardware, not before.
• The tray and the desktop window without a reinstallA PyPI install without the desktop extra has no upgrade path to the tray, the native window or a shortcut. All three should be a press in Settings, on the same pinned-wheel path that already fetches the speech stack.
• Agent reach without MCP, and an authenticated loopbackThe local API authenticates nothing today, which is the gate on everything else: identify session-originated requests server-side, then let an agent reach the daemon through the CLI as well as MCP, through the same authority matrix.
• Updates that only transfer what changedAn update already rewrites only the files that moved. It still ships the whole archive to do it.
• Git operations from the Git tabCommit with a message, manage the ignore file, and see failures plainly rather than dropping to a shell. The commit message is worth generating from the diff and the session record behind it.
• Session restore after a restart or a crashSessions already outlive the browser, the daemon and app updates. They should also survive the machine going down, and offer to reopen what was open.
• Hooks on swe-mux's own signalsHarness-neutral hooks that fire on PTY events, status transitions and the scan timeline, so a workflow built around them keeps working when you change agent CLI.
• Deeper code analysisJoining the structural graph to the behavioral record at the symbol level: cross-session conflict prediction, entity-level change facts, and stale-caller detection. swe-mux is the only observer that sees every parallel session's edits as they happen, which is what makes the first of those possible at all.
• A portable verification contractThe setup and verify scripts a worktree needs are already repository-neutral, and the land queue is worth nothing without them. Making that pair reachable and adoptable from inside any repository is the gap.
• Model routing in the assistantA small request should not cost a large model. Route simple asks cheaply and escalate the ones that need reasoning or several tool calls in a row.
• A shared browser surfaceGated on an authorization decision that has not been made. Today's previews are safe precisely because they cannot navigate anywhere and no agent can drive them; letting either change is a product decision about trust, not a feature.

Feature requests are GitHub Discussions in the Ideas category, and a thumbs-up on one is a vote. The issue tracker stays for bugs.

Describe the problem, not the fix. What you were trying to do, what you did instead, and what made it hard. A vote says what to look at first and nothing more; nothing is scheduled by being voted for.