Architecture

The parts of TurboDL and how the engine works.

The parts

shared engine      Rust library, no UI
turbodl            egui app
turbodl-cli        command line
turbodl-mcp        MCP server (rmcp)
extensions         Chrome and Firefox
profiles           default profile and JSON Schema

The engine

Module	Purpose
`engine/probe`	First request: size, byte ranges, name, ETag; network retries
`engine/plan`	Segments restored from the state or split from scratch
`engine/worker`	One connection per segment, batched writes (256 KiB), retries with growing waits
`engine/splitter`	Initial split and re-split of the largest remaining segment
`engine/finish`	Destination, state saved every 2 s, finalization, history
`resume`	`.part.turbodl` file without secrets, written atomically
`request` / `auth`	Request, `${VAR}` headers, Netscape cookies, Basic
`resolvers`	Google Drive
`stats`	`history.jsonl` history, aggregates and insights

Front ends get no message per received block: they read a snapshot at their own pace, which keeps the display smooth whatever the speed.

The desktop app

Module	Purpose
`queue`	Download list, waiting queue, persistence in `downloads.json`
`bridge`	Single instance and native messaging host for the extensions
`i18n`	Interface texts in English, French and Spanish; English is the fallback
`theme` + `icons`	Light/dark themes tinted by the accent color, Phosphor icon font
`update`	Version lookup, package download with the engine, checksum, file replacement with rollback
`ui`	Windows and dialogs

Technical choices

Rust: no garbage collector nor global lock, one native program per platform.
egui: immediate-mode interface rendered by the GPU, without an embedded browser.
rustls: TLS without depending on OpenSSL.