⚡ Bolt: [performance improvement]

.github/workflows/docs.yml

@@ -0,0 +1,49 @@
+name: Deploy documentation
+
+on:
+  push:
+    branches:
+      - master
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+concurrency:
+  group: pages
+  cancel-in-progress: false
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.x"
+
+      - name: Install documentation dependencies
+        run: python -m pip install --upgrade pip && python -m pip install -r requirements-docs.txt
+
+      - name: Build documentation
+        run: python -m mkdocs build --strict
+
+      - name: Configure GitHub Pages
+        uses: actions/configure-pages@v5
+
+      - name: Upload site artifact
+        uses: actions/upload-pages-artifact@v3
+        with:
+          path: site
+
+      - name: Deploy to GitHub Pages
+        id: deployment
+        uses: actions/deploy-pages@v4

.gitignore

@@ -1,89 +1,36 @@
-# ═══════════════════════════════════════════════════════════════════════════════
-# AETHER Project Gitignore
-# ═══════════════════════════════════════════════════════════════════════════════
-
 # Rust
-/target/
+/target
+/debug
+/release
 **/*.rs.bk
 Cargo.lock
-**/*.profraw
-**/*.profdata
-
-# Python
-__pycache__/
-*.py[cod]
-*$py.class
-*.so
-.Python
-build/
-develop-eggs/
-dist/
-downloads/
-eggs/
-.eggs/
-lib/
-lib64/
-parts/
-sdist/
-var/
-wheels/
-share/python-wheels/
-*.egg-info/
-.installed.cfg
-*.egg
-MANIFEST
 
-# Virtual Environment
-venv/
-.venv/
-env/
-.env/
+# Build artifacts
+build_err*.txt
+build_log*.txt
+check_output*.txt
+test_output.txt
+*.log
+*.out
+/site/
 
-# IDE/Editor
-.idea/
-.vscode/
+# Editors
+.vscode
+.idea
 *.swp
 *.swo
-*.sublime-project
-*.sublime-workspace
 
-# OS Generated Files
+# OS
 .DS_Store
-.DS_Store?
-._*
-.Spotlight-V100
-.Trashes
-ehthumbs.db
 Thumbs.db
 
-# Logs & Diagnostics
-*.log
-*.txt
-*.tmp
-build_err*.txt
-build_log*.txt
-check_*.log
-error.log
-test_*.log
-
-# Build Artifacts
-*.bin
-*.img
-*.iso
-*.iso.bin
-*.iso.lock
-
-# Agents / AI
-agents/
-.gemini/
-.antigravity/
-
-# Archives
-*.zip
-*.tar
-*.tar.gz
-*.rar
-
-# Docker
-.dockerignore
-
+# Python
+__pycache__
+*.pyc
+.venv
+venv
+test_perf/
+
+# Agent scratch
+.jules/
+jules/

.jules/bolt.md

@@ -0,0 +1,12 @@
+## 2026-07-06 - Tensor metadata cloning in autograd
+**Learning:** In reverse-mode autograd passes, cloning `Option<Tensor>` or passing references triggers unnecessary heap allocations for tensor metadata (shape/strides), even though the underlying data is reference-counted.
+**Action:** Use `Option::take()` to acquire ownership of gradients during the backward pass, and pass tensors by value to `accumulate_grad` to eliminate metadata clones.
+## 2026-07-14 - Optimizing Tensor allocations in linear algebra
+**Learning:** High-level tensor operations like `.sub()` and `.map()` during gradient calculations trigger costly intermediate heap allocations for both data and metadata.
+**Action:** Use single-pass iterators (`.iter().zip().map().collect()`) directly over the borrowed data arrays and consume the resulting vector with `Tensor::from_vec()` to avoid redundant O(N) slice allocations.
+## 2026-07-25 - Tensor metadata cloning in MLP forward passes
+**Learning:** In `aether-core::ml::neural`, cloning the `input` tensor in `MLP::forward` before passing its reference to the first layer's `forward` method triggers an unnecessary heap allocation for tensor metadata and an `Rc` increment.
+**Action:** Extract the first layer using `self.layers.iter_mut()` to pass the initial `input` as a `&Tensor` reference directly, as subsequent layers naturally consume the output of the previous layer.
+## 2026-08-03 - [Optimize DenseLayer backward pass]
+**Learning:** In aether-core, DenseLayer::backward is optimized by using Option::take() on last_z and last_input instead of cloning them. This takes ownership, avoiding unnecessary heap allocations for shape and stride metadata while preventing mutable borrow conflicts during backpropagation.
+**Action:** When working with cached tensor state during backpropagation, use Option::take() to transfer ownership rather than cloning, provided the state isn't needed again before the next forward pass.

Aether.lean

@@ -0,0 +1,6 @@
+import Aether.Lexer
+import Aether.Core
+import Aether.Static
+import Aether.Parser
+import Aether.VM
+import Aether.Pipeline