Tauri Audit 2026

Tauri audit (2026-05-11)

1) Executive summary

Status (2026-05-11 update): ADR 037 implementation is in progress on main: vox compile --target desktop|mobile-* emits a real Tauri 2 workspace under target/generated/ (src-tauri/, tauri.conf.json); native-binary keeps the Axum + embedded SPA path. apps/vox-mental-tracker consumes vox-tauri-sherpa-guest instead of Capacitor. Historical evidence below is retained; treat numbered findings as a time-capsule unless the surrounding paragraph says current.

Canonical decision: ADR 037 — Tauri Convergence. Execution checklist: Tauri convergence migration plan (2026-Q2).

Prior audit conclusion (pre-implementation):

Tauri was previously a manifest-hint layer, not a full build/runtime dependency; mobile dev/e2e ran through Capacitor. That split is retired for vox-mental-tracker and the codegen desktop/mobile shell.

2) Current-state audit (what code does today)

2.1 Tauri in Rust deps: absent from generated runtime

vox-tauri-codegen itself is serialization + file emission (anyhow, serde*, vox-compiler), with no Tauri runtime crates (crates/vox-tauri-codegen/Cargo.toml:9-14).
Generated backend Cargo.toml includes Axum/Tokio/rust-embed and Vox crates, but no tauri or tauri-build (crates/vox-codegen/src/codegen_rust/emit/mod.rs:153-173).

2.2 `vox compile` behavior by target

native-binary calls bundle app mode only (crates/vox-cli/src/commands/compile.rs:56-64).
desktop, mobile-android, mobile-ios call bundle app mode, then emit Tauri packaging hints (crates/vox-cli/src/commands/compile.rs:66-76, 251-272).
server is explicitly redirected to deploy workflows, not packaging (crates/vox-cli/src/commands/compile.rs:85-89).

2.3 What bundle/build produce

vox bundle app path runs build pipeline + Vite/pnpm + cargo build for vox_generated_app (crates/vox-cli/src/commands/bundle.rs:86-160, 257-298).
Generated runtime binds Axum on 127.0.0.1 and serves embedded/static content (crates/vox-codegen/src/codegen_rust/emit/http.rs:304-410).
vox build mobile path currently calls Capacitor sync (npx cap sync) when target is ios/android (crates/vox-cli/src/commands/build.rs:345-360).

2.4 What `vox-tauri-codegen` emits

Emits tauri-packaging/tauri.conf.json + README.md; optionally emits runtime-capabilities.projection.json when capability SSOT exists (crates/vox-tauri-codegen/src/lib.rs:195-219, 221-246).
Uses find_contracts_repo_root to walk upward for contracts/capability/runtime-capabilities.v1.yaml (crates/vox-tauri-codegen/src/lib.rs:55-68).
README explicitly states consumers still run cargo tauri build in their own Tauri workspace (crates/vox-tauri-codegen/src/lib.rs:233-237).

2.5 CI behavior (compile and mobile)

Compile-matrix smoke builds vox-cli, runs vox ci compile-matrix, and Linux also smoke-tests compile-suite native-binary; no cargo tauri build step exists (.github/workflows/compile-matrix.yml:33-37, 45-55).
Android e2e path is Capacitor/Gradle (pnpm build && npx cap sync android, gradlew assembleDebug) (.github/workflows/mobile-e2e-android.yml:31-43).

3) Claimed-state audit (what docs currently claim)

Packaging SSOT says desktop/mobile installers are produced via Tauri 2 scaffolding (docs/src/architecture/vox-application-packaging-ssot-2026.md:59-63).
Journey C claims vox compile --target mobile-android|mobile-ios produces install artifacts when tooling is present (docs/src/architecture/vox-application-packaging-ssot-2026.md:41-43).
GUI roadmap and ADR-024 position the dashboard/web direction around Axum-served SPA and explicitly rejected Tauri for that dashboard host (docs/src/architecture/vox-gui-native-roadmap-2026.md:22-24; docs/src/adr/024-dashboard-axum-spa.md:26-29, 35-38).

Net: language-level roadmap and dashboard architecture are Axum/web-first, while packaging SSOT currently overstates live Tauri execution.

4) Tauri 2 mobile feasibility for Vox apps

4.1 What is already feasible

For non-ASR-heavy mobile GUI apps, Tauri 2 mobile can satisfy the same practical surface area currently used through Capacitor:

Filesystem, clipboard, notifications, deep-link, HTTP, and shell-level integration (Tauri 2 plugin model).
Standard webview host model (Android/iOS webviews).
Packaging and permission model that can consume capability maps.

4.2 Concrete blocker found in this repo

vox-mental-tracker documents why it remains on Capacitor:

Packaging note says migration waits for Tauri-native Sherpa equivalent (apps/vox-mental-tracker/README.md:5).
App deps/scripts are explicitly Capacitor-first (apps/vox-mental-tracker/package.json:8-16, 23, 40-42).

This is the key gating item for full mobile unification in this codebase.

5) Build-time and CI cost picture

Today

No Tauri Rust compile in generated backend path.
Compile matrix validates command wiring and one Linux native-binary workspace smoke only (.github/workflows/compile-matrix.yml:33-37).
Mobile CI cost currently sits in Node+Capacitor+Gradle (.github/workflows/mobile-e2e-android.yml:31-43).

If real Tauri builder were adopted

Generated app crate would need Tauri runtime/build deps (not present today in emit_cargo_toml) (crates/vox-codegen/src/codegen_rust/emit/mod.rs:153-173).
CI would need explicit platform Tauri build lanes (desktop and/or mobile), adding webview/toolchain setup and longer cold build/link stages.

If staying hint-only

Current compile performance profile remains mostly unchanged; only docs/validation surface changes.

6) Strategic option matrix (cost/risk framing)

Option	Engineering cost	CI impact	Runtime alignment	Main risk
Adopt Tauri desktop+mobile now	High	High	Strong single shell	Tauri mobile maturity + Sherpa port effort
Tauri desktop, Capacitor mobile	Medium	Medium	Split pipeline	Ongoing dual-maintenance burden
Keep hybrid/hints, clean seams	Low-medium	Low	Honest to current reality	Defers full installer unification
Retire Tauri hints entirely	Low	Low	Simplifies now	Loses forward path to installer-native lane

7) Free-wins retirement candidates (independent of strategy)

tauri_stub seam appears unused:
- Module + re-export exist (crates/vox-codegen/src/codegen_rust/emit/tauri_stub.rs:1-12, crates/vox-codegen/src/codegen_rust/emit/mod.rs:18, 30).
mobile_emit hard-imports @tauri-apps/api/* when mobile hooks exist (crates/vox-codegen/src/codegen_ts/mobile_emit.rs:26-31) but templates/scaffolds do not provision those packages by default in this repo’s common generation path.
init prints a target/generated/tauri-packaging next-step for mobile templates, but current emission there writes to project-root tauri-packaging/ (crates/vox-cli/src/commands/init.rs:33-35, 94-103).
Packaging SSOT wording should be made explicitly current-state or explicitly planned-state to avoid drift (docs/src/architecture/vox-application-packaging-ssot-2026.md:59-63).
Add a generation contract test around mobile.ts import assumptions before changing strategy.

8) Recommendation framing (advisory)

Given the stated goal (“single pipeline we have to maintain mobile and desktop”):

If on-device Sherpa transcription remains mandatory in the near term, either:
- Keep hybrid while scheduling Sherpa plugin port to Tauri mobile, then converge, or
- Accept temporary split and clearly timebox it.
If Sherpa dependency can be deferred/substituted, full Tauri convergence becomes materially simpler.

This is a product/engineering trade-off decision, not a tooling mystery:

most non-ASR mobile capabilities are not the blocker,
native Sherpa bridge shape is the blocker in this repo.

9) Decision criteria for follow-up ADR

The follow-up ADR should explicitly answer:

Is temporary dual-shell operation acceptable, and for how long?
Who owns vox-sherpa-transcribe Tauri porting, and what acceptance tests gate completion?
Should native-binary remain a first-class Axum lane even after Tauri adoption?
Should capability projection remain merge-hint JSON or become directly consumed by generated Tauri config/capability files?
What CI lanes become required (desktop only vs full mobile)?

10) References

In-repo

External context (for mobile maturity and plugin availability checks)

Notes

This audit intentionally records what is implemented now and separates that from strategic direction decisions.
No code or pipeline behavior is changed by this document.
Resolved by: ADR 037 — Tauri Convergence and Tauri convergence migration plan (2026-Q2).