Roadmap

Roadmap

Tracked on the project board and in issues.

Phase 0 — Scaffold & repo hygiene ✅

Complete — merged in #20, milestone closed, CI green on main.

• Next.js 16 static export inside a Tauri v2 shell
• librqbit v9 embedded, DHT bootstrapping, UPnP forwarding, persistence
• get_core_status proving the full IPC path end to end
• Dark landing page showing live engine telemetry
• Apache-2.0, README, CONTRIBUTING, issue/PR templates, Dependabot
• CI: fmt, clippy, tsc, ESLint, Prettier, Vitest, cargo test, audits
• 5 Rust unit tests, 6 integration tests (engine + IPC), 20 frontend tests
• Three security advisories found and fixed by CI before the first merge

Phase 1 — Core torrent lifecycle (MVP) 🚧

The phase that makes Flume usable. The core loop works: telemetry is event-based, and torrents can be added by magnet or file with a select-files-first flow, then paused, resumed, and removed.

• Add via magnet link ✅ (#3)
• Add via .torrent file picker ✅ (#4)
• Torrent list: progress, speeds, ETA, peers, ratio; pause/resume/remove ✅ (#5)
• Per-torrent file tree with selective download ✅ (#6)
• Settings with persistence ✅ (#7)
• Resume correctly across restarts ✅ (#8)
• Investigate Windows file-locking and seeding (#9)

Also in this phase: replace the Phase 0 polling hook with backend-pushed, batched events before the torrent count grows.

Phase 2 — Polish & platform integration ✅

Complete. Milestone closed 5/5.

• Theming and the visual design pass ✅ (#10)
• Per-torrent detail view with piece heatmap ✅ (#11)
• Magnet protocol association and single instance ✅ (#12)
• Notifications and system tray ✅ (#13)
• Keyboard shortcuts and accessibility baseline ✅ (#14)

Phase 3 — Hardening & distribution 🚧

• Release pipeline for all four package formats ✅ (#15)
• Performance validation with 10+ torrents ✅ (#17)
• Signing, notarization, and troubleshooting docs ✅ (#18)
• Release build blocked by a proc-macro failure ✅ (#22)

Still open, all blocked on something outside the code:

• Sequential download (#16) — not implementable against librqbit v9; FilePriorities is pub(crate) and no priority setter exists. Awaiting a product decision.
• GTK3 advisories (#21) — upstream in Tauri's Linux backend; resolves when it moves off GTK3.
• TypeScript 7 (#28) — blocked until typescript-eslint supports it.

Measured performance

With 15 torrents on macOS:

Metric	Value	Budget
telemetry() per call	171 µs	1,000,000 µs (1 Hz tick)
Serialized payload	5,345 bytes	—
Detail + files query	2.2 µs	500,000 µs (2 Hz panel)

The payload figure is the one to watch: it should scale with torrent count, never with piece count or file size.

Known limitations

• Windows and Linux unverified locally. Developed on macOS 27. CI builds for all platforms, but first-run behaviour needs a manual pass.
• Polling, not events. The Phase 0 status hook polls at 1 Hz. Fine for one status card, wrong for a torrent list.
• No published builds. Until Phase 3, build from source.
• Magnet association is untested on macOS and Windows. The OS registration lives in the installed bundle, so it cannot be exercised by tauri dev — see Platform-Notes. Blocked behind the release pipeline (#15), which is itself blocked by #22.

Out of scope

Flume deliberately does not do cryptocurrency, chat, RSS automation, or paid features. It also will not ship a second CLI — rqbit already exists.