Refactor: Crate Extraction

.github/workflows/ci.yml

@@ -22,22 +22,56 @@ jobs:
       - name: Cache cargo
         uses: Swatinem/rust-cache@v2
 
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 22
+
+      - name: Install UI dependencies
+        working-directory: crates/promptforge-ws-server/ui
+        run: npm ci
+
       - name: Format
         run: cargo fmt --all --check
 
+      # promptforge-gateway and promptforge-transcribe are excluded from the
+      # blanket commands: --all-features would enable workshop-cuda and the
+      # transcription engine's cuda feature, which need a CUDA Toolkit these
+      # runners do not have. Their non-CUDA surfaces are exercised by the
+      # dedicated steps below.
       - name: Clippy
-        run: cargo clippy --workspace --exclude promptforge-ws --exclude promptforge-ws-server --all-targets --all-features -- -D warnings
+        run: cargo clippy --workspace --exclude promptforge-ws --exclude promptforge-ws-server --exclude promptforge-desktop-shell --exclude promptforge-gateway --exclude promptforge-transcribe --all-targets --all-features -- -D warnings
 
       - name: Test
-        run: cargo test --locked --workspace --exclude promptforge-ws --exclude promptforge-ws-server --all-features
+        run: cargo test --locked --workspace --exclude promptforge-ws --exclude promptforge-ws-server --exclude promptforge-desktop-shell --exclude promptforge-gateway --exclude promptforge-transcribe --all-features
 
       - name: Doctests
-        run: cargo test --workspace --exclude promptforge-ws --exclude promptforge-ws-server --doc
+        run: cargo test --workspace --exclude promptforge-ws --exclude promptforge-ws-server --exclude promptforge-desktop-shell --exclude promptforge-gateway --exclude promptforge-transcribe --doc
 
       - name: Docs
         env:
           RUSTDOCFLAGS: -D warnings
-        run: cargo doc --workspace --no-deps --all-features --exclude promptforge-ws --exclude promptforge-ws-server
+        run: cargo doc --workspace --no-deps --all-features --exclude promptforge-ws --exclude promptforge-ws-server --exclude promptforge-desktop-shell --exclude promptforge-gateway --exclude promptforge-transcribe
+
+      # The transcription engine without its cuda feature: whisper then builds
+      # its CPU backend, which needs no toolkit.
+      - name: Clippy (transcribe)
+        run: cargo clippy -p promptforge-transcribe --all-targets -- -D warnings
+
+      - name: Test (transcribe)
+        run: cargo test --locked -p promptforge-transcribe
+
+      - name: Docs (transcribe)
+        env:
+          RUSTDOCFLAGS: -D warnings
+        run: cargo doc --no-deps -p promptforge-transcribe
+
+      # The workshop feature hosts the ws-server UI; its build script bundles
+      # the UI with esbuild, which needs the node_modules installed above.
+      - name: Clippy (gateway)
+        run: cargo clippy -p promptforge-gateway --all-targets --features workshop -- -D warnings
+
+      - name: Test (gateway)
+        run: cargo test --locked -p promptforge-gateway --features workshop
 
   check-workshop:
     runs-on: windows-latest
@@ -63,25 +97,45 @@ jobs:
       # feature, and windows-latest has no CUDA toolkit; CI builds the
       # CPU-only whisper path instead.
       - name: Clippy (workshop)
-        run: cargo clippy -p promptforge-ws -p promptforge-ws-server --all-targets --no-default-features -- -D warnings
+        run: cargo clippy -p promptforge-ws -p promptforge-ws-server -p promptforge-desktop-shell --all-targets --no-default-features -- -D warnings
 
       - name: Test (workshop)
-        run: cargo test --locked -p promptforge-ws -p promptforge-ws-server --no-default-features
+        run: cargo test --locked -p promptforge-ws -p promptforge-ws-server -p promptforge-desktop-shell --no-default-features
 
   msrv:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
 
       - uses: dtolnay/rust-toolchain@1.89.0
+        with:
+          components: rustfmt
 
       - name: Cache cargo
         uses: Swatinem/rust-cache@v2
 
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 22
+
+      - name: Install UI dependencies
+        working-directory: crates/promptforge-ws-server/ui
+        run: npm ci
+
       - name: Build and test on MSRV
         run: |
-          cargo build --locked --workspace --exclude promptforge-ws --exclude promptforge-ws-server --all-features
-          cargo test --locked --workspace --exclude promptforge-ws --exclude promptforge-ws-server --all-features
+          cargo build --locked --workspace --exclude promptforge-ws --exclude promptforge-ws-server --exclude promptforge-desktop-shell --exclude promptforge-gateway --exclude promptforge-transcribe --all-features
+          cargo test --locked --workspace --exclude promptforge-ws --exclude promptforge-ws-server --exclude promptforge-desktop-shell --exclude promptforge-gateway --exclude promptforge-transcribe --all-features
+
+      - name: Build and test gateway on MSRV
+        run: |
+          cargo build --locked -p promptforge-gateway --features workshop
+          cargo test --locked -p promptforge-gateway --features workshop
+
+      - name: Build and test transcribe on MSRV
+        run: |
+          cargo build --locked -p promptforge-transcribe
+          cargo test --locked -p promptforge-transcribe
 
   supply-chain:
     runs-on: ubuntu-latest

.github/workflows/cuda.yml

@@ -0,0 +1,34 @@
+name: CUDA
+
+on:
+  workflow_dispatch:
+  # Nightly CUDA build and live test. Uncomment once a self-hosted runner
+  # with the `windows` and `cuda` labels is registered.
+  # schedule:
+  #   - cron: '0 6 * * *'
+
+jobs:
+  cuda:
+    runs-on: [self-hosted, windows, cuda]
+    timeout-minutes: 120
+    steps:
+      # submodules: the CUDA build compiles the pinned llama.cpp submodule.
+      - uses: actions/checkout@v4
+        with:
+          submodules: true
+
+      - uses: dtolnay/rust-toolchain@stable
+
+      - name: Cache cargo
+        uses: Swatinem/rust-cache@v2
+
+      - name: Build CUDA bundle
+        run: cargo build --locked -p promptforge-ws
+
+      - name: Test (llama-cuda)
+        run: cargo test --locked -p promptforge-gateway --features llama-cuda
+
+      - name: Live CUDA test
+        env:
+          PROMPTFORGE_LIVE_CUDA: "1"
+        run: cargo test --locked -p promptforge-gateway --features llama-cuda -- --ignored live_cuda --nocapture

.gitignore

@@ -6,7 +6,7 @@
 /guide/book/
 *.env
 # Voice test fixtures, downloaded out of band (see design-promptforge-ws-1.md).
-/crates/promptforge-ws-server/tests/fixtures/
+/crates/promptforge-transcribe/tests/fixtures/
 # UI build pipeline: npm install target and esbuild output (rebuilt by build.rs).
 /crates/promptforge-ws-server/ui/node_modules/
 /crates/promptforge-ws-server/ui/dist/

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "third_party/llama.cpp"]
+	path = third_party/llama.cpp
+	url = https://github.com/ggml-org/llama.cpp.git

AGENTS.md

@@ -13,6 +13,7 @@ This rule outranks every other rule here. Before you add a frontmatter field, a
 - Each crate's own AGENTS.md binds its subtree. Read the ones governing the paths you touch before writing or reviewing code.
 - The existing test suite stays green and intact during refactors: fix forward; never rewrite a test to make it pass.
 - Library and serve paths never call `process::exit` or install process-global state; failures return through the spawn handshake.
+- Runtime and serve paths never compile native dependencies or invoke compilers or build tools. Native compilation belongs to the Cargo build or packaging process; runtime may only verify, stage, and launch build-produced artifacts.
 - Do NOT look at files outside this repo for reference.
 - The plan is the spec. Work from the plan and AGENTS.md only.
 

Cargo.toml

@@ -11,10 +11,15 @@ repository = "https://github.com/cppalliance/promptforge"
 
 [workspace.dependencies]
 promptforge-core = { path = "crates/promptforge-core", version = "0.1.0" }
+promptforge-desktop-shell = { path = "crates/promptforge-desktop-shell", version = "0.1.0" }
 promptforge-gateway = { path = "crates/promptforge-gateway", version = "0.1.0" }
+promptforge-gateway-build = { path = "crates/promptforge-gateway-build", version = "0.1.0" }
 promptforge-gateway-config = { path = "crates/promptforge-gateway-config", version = "0.1.0" }
 promptforge-webfetch = { path = "crates/promptforge-webfetch", version = "0.1.0" }
 promptforge-tool-picker = { path = "crates/promptforge-tool-picker", version = "0.1.0" }
+promptforge-tools = { path = "crates/promptforge-tools", version = "0.1.0" }
+promptforge-transcribe = { path = "crates/promptforge-transcribe", version = "0.1.0" }
+promptforge-web-search = { path = "crates/promptforge-web-search", version = "0.1.0" }
 promptforge-ws-server = { path = "crates/promptforge-ws-server", version = "0.1.0" }
 pulldown-cmark = "0.12"
 serde = { version = "1", features = ["derive"] }
@@ -85,10 +90,12 @@ tar = { version = "0.4.46", default-features = false }
 zip = { version = "8.6.0", default-features = false, features = ["deflate-flate2"] }
 indicatif = "0.18"
 # Safe bindings to whisper.cpp; the C++ core is built from source by
-# whisper-rs-sys (cmake + MSVC required). Enable the wb-server's "cuda"
-# feature to build with GPU acceleration (needs the CUDA Toolkit).
+# whisper-rs-sys (cmake + MSVC required). Enable promptforge-transcribe's
+# "cuda" feature (via the ws-server's "voice-cuda") to build with GPU
+# acceleration (needs the CUDA Toolkit).
 whisper-rs = "0.16"
-# WAV parsing for the voice test fixtures; dev-dependency of the wb-server.
+# WAV parsing for the voice test fixtures; optional dependency of
+# promptforge-transcribe's test-fixtures feature.
 hound = "3.5"
 socket2 = { version = "0.6", features = ["all"] }
 # wry 0.56 pairs with tao 0.36 (wry's own dev-dependency constraint); tao 0.37