Coding Owl is archived. It did not fulfill its core loop, so rather than pivoting or refactoring it, I am building something new from scratch with the lessons learned. Why Coding Owl is archived

ADR-0009

Desktop app is Wails with a React frontend

Context

Coding Owl ships a desktop dashboard alongside the CLI, in the same release. It wants to feel native on macOS, live in the monorepo, and reuse the Go domain code that already exists for the daemon.

Decision

Wails - Go backend, React/TypeScript frontend - as a separate build target from the owl binary.

The GUI is a pure view. Every action it can perform exists as a daemon RPC that the CLI can also call; internal/daemon remains the single source of truth. The app is a client of the daemon like any other.

Visual direction: the same design as codingowl.dev. A neutral-50 ground, white where a surface is needed, hairline rules to divide, Geist for text and Geist Mono for anything mechanical, and one sky blue spent sparingly. Colour says one of four things and nothing else - blue is in motion, near-black is waiting for you, rose is something that went wrong, grey is at rest - so a coloured thing on screen always means something.

Amended 2026-09-17 (#36). This originally read “liquid glass, night-sky and navy blues”, and the app shipped that way in #9. The design review found the glass fought the content: a dashboard is mostly small text in dense tables, and blur, translucency and saturated navy all work against reading it. The site had already moved to the quieter design; the app follows it, so there is one look to maintain rather than two. The theme is two files - frontend/src/theme/tokens.css for the values and frontend/src/style.css for what uses them - so the next change is as small as this one was.

Superseded by ADR-0029. This ADR originally fixed a day-one feature set for the app. There is no longer one: the core loop ships first in the daemon, and features land across daemon, API and desktop app together thereafter. Diff review remains explicitly out - that is what ADR-0007’s branches are for.

Consequences

  • The Go side of the app reuses internal/client, so there is no second API surface and no risk of the GUI growing behaviour the CLI cannot reach.
  • Node and pnpm join the build for the frontend. pnpm is the frontend’s one package manager: its version is pinned in package.json, its lockfile is committed, and every build - local, CI, release - installs from that lockfile alone. (Amended 2026-09-14: npm until then.)
  • Because the frontend talks to Go through Wails bindings rather than to the daemon directly, no TypeScript client is generated from the protobufs. This is worth stating because it removes what looks like an obvious argument for gRPC in ADR-0004 - the real argument there is streaming.
  • Shipping a macOS cask eventually means codesigning and notarization, which needs an Apple Developer Program membership. For now the app is ad-hoc signed and not notarised, following the maintainer’s Reviewdeck cask: the cask clears the quarantine flag itself in a postflight step. Notarisation is deferred, and with it the only administrative dependency a first release had.

Alternatives considered

Electron. Heavier, and no reuse of the Go domain code.

Tauri. Good product, but introduces Rust as a second systems language in a solo project.

An HTTP dashboard served by the daemon with the frontend embedded via Go embed, plus a thin menubar wrapper later. Retained as the fallback if Wails proves heavy in practice.