add multi-platform inference framework

.gitattributes

@@ -0,0 +1,24 @@
+# House rule: all text is stored and checked out with LF.
+# (Batch wrappers are the only exception -- cmd.exe needs CRLF.)
+* text=auto eol=lf
+*.bat text eol=crlf
+*.cmd text eol=crlf
+
+# Binaries: never touch line endings, never treat as text
+*.so binary
+*.a binary
+*.dll binary
+*.exe binary
+*.png binary
+*.jpg binary
+*.jpeg binary
+*.ico binary
+*.icns binary
+*.jar binary
+*.zip binary
+*.bin binary
+*.gguf binary
+*.safetensors binary
+*.e0b binary
+# repo-specific addition: the technical report lives at paper/main.pdf
+*.pdf binary

.github/workflows/ci.yml

@@ -9,6 +9,9 @@ jobs:
   test-macos:
     name: Unit tests (macOS + MLX)
     runs-on: macos-14
+    defaults:
+      run:
+        working-directory: python
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
@@ -22,6 +25,9 @@ jobs:
   hygiene:
     name: Repo hygiene (pure Python, no MLX)
     runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: python
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5

.gitignore

@@ -13,9 +13,21 @@ dist/
 # Trained adapter artifacts (never committed; see README)
 artifacts/
 
+# Downloaded model checkpoints (never committed; see README quick start).
+# Anchored so the source package python/src/edge0/models/ is not matched.
+/models/
+/python/models/
+
+# Platform build areas materialized by the engine/vendor scripts (never commit):
+# isolated patch-replay worktrees, the pinned llama.cpp checkout, and the
+# local engine-depot bootstrap fallback (see windows/scripts/vendor-build.ps1).
+/wt/
+/vendor/
+.edge0/
+
 # macOS
 .DS_Store
 
 # Bench / runtime logs
 *.log
-scripts/alignment_results.json
+python/scripts/alignment_results.json

NOTICE

@@ -4,18 +4,32 @@ edge0
 This project bundles or adapts the following third-party works:
 
 mlx-lm (MIT License)
-  - `src/edge0/backends/mlx/_impl/qwen3_next.py` is vendored from mlx-lm
+  - `python/src/edge0/backends/mlx/_impl/qwen3_next.py` is vendored from mlx-lm
     0.31.0 (https://github.com/ml-explore/mlx-lm), with additive hooks for
     prerouter routing and layer callbacks.  Non-prerouter paths are kept
     byte-identical to the upstream file.
   - mlx-lm is used at runtime for checkpoint / tokenizer loading.
   Copyright (c) 2023-2024 Apple Inc. and the mlx-lm contributors.
 
 bailing_hybrid (Apache License 2.0)
-  - `src/edge0/backends/mlx/_impl/bailing_hybrid.py` is vendored from the
+  - `python/src/edge0/backends/mlx/_impl/bailing_hybrid.py` is vendored from the
     Ling 3.0 MLX implementation (rapid-mlx / Ling-3.0-tiny, Apache-2.0), with
     prerouter hook identifiers renamed for consistency with the framework.
   Copyright (c) the bailing / Ling MLX contributors.
 
+llama.cpp (MIT License)
+  - The Android and Windows engines build their inference core from a
+    pristine upstream checkout of llama.cpp
+    (https://github.com/ggml-org/llama.cpp), pinned at commit 7ab4ee7
+    (release b11100) via `vendor.llama.pin` and materialized into the
+    gitignored `vendor/llama.cpp` by the platform build scripts.
+  - It is never patched in place: all modifications ship as replayable
+    patch bands under `patches/llama.cpp/{common,android,windows}` and are
+    replayed into an isolated build worktree at compile time.
+  - ggml (Apache License 2.0) is developed within the same upstream tree.
+  - Not vendored as files in this repo; built from source by the consumer
+    (see `android/NOTICE` and `windows/NOTICE` for the per-platform detail).
+  Copyright (c) 2023-2024 The ggml authors.
+
 All other files are Copyright (c) 2025 edge0 contributors, licensed under
 the Apache License, Version 2.0 (see LICENSE).