Project

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.

  • 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.