diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..c101618
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,32 @@
+# Python
+.venv/
+venv/
+__pycache__/
+*.pyc
+*.pyo
+*.egg-info/
+build/
+dist/
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+
+# Data / models / recordings (large, not source)
+*.mp4
+*.mkv
+*.avi
+data/
+datasets/
+checkpoints/
+*.pt
+*.pth
+*.onnx
+
+# OS
+.DS_Store
+Thumbs.db
+
+# Editors
+.vscode/
+.idea/
+*.swp
diff --git a/CLAUDE.md b/CLAUDE.md
new file mode 100644
index 0000000..f7c2856
--- /dev/null
+++ b/CLAUDE.md
@@ -0,0 +1,162 @@
+# SpelunkAI — Project Definition
+
+- **Project name:** SpelunkAI
+- **Git repository:** https://git.raggonium.net/Jonas/spelunkai
+- **Commit convention:** Every commit message must include the name of the Claude model that made it (i.e. the current model name in use at the time of the commit).
+
+## 1. Project Goal
+
+Build a multi-stage AI system that learns to play *Spelunky Classic HD* by combining:
+
+1. Trained perception models (custom, anchor-free CNN detectors) for entity/hazard/item recognition
+2. Hard, fast, deterministic high-level logic classifiers for HUD state (health, gold, bombs, ropes, etc.)
+3. A control agent that first imitates human input (behavior cloning) and later learns autonomously (reinforcement learning)
+
+The entire perception → reasoning → action loop must fit inside a **66ms compute budget per tick** (~15 Hz), since some detection models may need to run slower than others depending on what they detect. Efficiency is a first-class design constraint, not an afterthought — every architecture decision should be made with this budget in mind.
+
+A distant, non-blocking stretch goal is to eventually port the trained models to run on an FPGA (HDMI capture in, USB/game input out). This does not affect current implementation priorities but should inform model size/efficiency choices where it's free to do so.
+
+---
+
+## 2. System Overview
+
+### 2.1 Hardware / Machines
+
+| Machine | Role | Notes |
+|---|---|---|
+| New recording PC | Gameplay recording | Ubuntu 24 Desktop, dedicated to running Spelunky Classic HD fullscreen and recording sessions |
+| `jai` (existing training server) | Labeling backend, model training, inference, control agent | Ubuntu 24.04, Ryzen 7 5800X, 64GB RAM, RTX 3070 Ti (~7.45 GiB VRAM) |
+
+### 2.2 High-Level Data Flow
+
+```
+[Recording PC]
+  Spelunky Classic HD (fullscreen, 1280x720)
+        │
+        ▼
+  Recording Script (30fps video + input log)
+        │
+        ▼
+  Lossless MP4 + input log file
+        │
+        ▼ (transfer to jai)
+[jai — Labeling & Training]
+  Web-based Labeling Tool
+   ├─ Set: Enemy      (Main: Enemy → Sub: Bat, Snake, ...)
+   ├─ Set: Items
+   ├─ Set: Traps
+   └─ Set: ... (ad-hoc, extensible)
+        │
+        ▼
+  Per-set training pipeline (anchor-free CNN detector)
+        │
+        ▼
+  Trained detection models + KPI evaluation
+        │
+        ▼
+[jai — Runtime / Inference Loop, budget: 66ms/tick]
+  Screen Capture
+        │
+        ▼
+  Parallel Detection Models (entities, per-set)
+        │
+        ▼
+  High-Level Logic Classifiers (health, gold, bombs, ropes, ... — fixed UI regions)
+        │
+        ▼
+  State Vector Builder
+   (entity list + relative positions, HUD status, downsampled raw frame)
+        │
+        ▼
+  Control Agent (behavior cloning → later RL)
+        │
+        ▼
+  Input Execution (keyboard)
+```
+
+---
+
+## 3. Component Specifications
+
+### 3.1 Recording Tool
+
+- **Platform:** Ubuntu 24 Desktop (new dedicated machine)
+- **Display:** Spelunky Classic HD run in fullscreen at 1280x720
+- **Input device:** Keyboard
+- **Video output:** Lossless MP4, 30 fps
+- **Input log:** Separate log file, input events mapped exactly to the 30fps video timeline (frame index + timestamp + key state)
+- **Requirements:**
+  - Frame-accurate sync between video frames and logged input state
+  - Output format must be easy to slice into individual frames for labeling later
+
+### 3.2 Web-Based Labeling Tool
+
+- **Bounding boxes only** (no segmentation/polygons needed)
+- **Hierarchical, ad-hoc class system:** Main class → Sub class (e.g. Main: `Enemy` → Sub: `Bat`, `Snake`, ...). Users can create new main/sub classes on the fly without a schema migration.
+- **Multiple independent label sets**, each targeting a specific model (e.g. `Enemy`, `Items`, `Traps`, `PositiveDetection`, ...). Frames may be shared across sets, but labels are tracked per set.
+- **Multi-user support**
+- **Simple frontend** — functionality over polish
+- **All labels must be editable** (move, resize, reclass, delete) at any time
+- **Active learning workflow:**
+  1. Manually label a small seed set (e.g. ~20 instances of a class)
+  2. Train an initial model on the seed set
+  3. Run inference on a larger new batch (e.g. ~100 frames) to auto-label
+  4. Review/correct the auto-labels in the UI
+  5. Promote to a new dataset version (versioned datasets, e.g. `enemy-v1`, `enemy-v2`, ...) for reproducibility
+- **Data model must support:** sets, hierarchical classes, dataset versions, per-image/per-set label state (unlabeled / auto-labeled / reviewed)
+
+### 3.3 Detection Models
+
+- **Approach:** Anchor-free, center-heatmap-based detector (CenterNet-style) — NOT YOLO/anchor-based, and NOT a two-stage (R-CNN-style) detector.
+  - Rationale: sprites have fairly consistent size/shape, ad-hoc new sub-classes shouldn't require anchor redesign, and heatmap peaks handle multiple same-class objects near each other well (e.g. several bats in frame).
+  - Model predicts: a center-point heatmap per class + direct width/height regression from each detected center.
+- **One model per label set** (Enemy, Items, Traps, ...), trained and run independently
+- **Custom, small/efficient CNN architectures** — sized with the 66ms total-pipeline budget in mind. Multiple detection models must be able to run in parallel (or fast serial) on the RTX 3070 Ti.
+- **Inference rate:** 15 fps baseline per model, slower where acceptable depending on what's being detected
+
+### 3.4 KPIs / Evaluation
+
+- **Phase 1:** Standard detection metrics evaluated against held-out labeled data (precision, recall, and an mAP-equivalent adapted for the heatmap/center-point formulation)
+- **Phase 2 (later):** Ground-truth validation via direct game-code output / memory reading (approach not yet researched — open item, see Roadmap)
+
+### 3.5 High-Level Logic Classifiers
+
+- Covers HUD/status information: health, gold count, bomb count, rope count, level, etc.
+- Since these UI elements sit at **fixed screen positions**, use small, dedicated classifiers per fixed UI region rather than generic OCR — faster and more deterministic.
+- These feed directly into the state vector as structured status values.
+
+### 3.6 Control Agent
+
+- **Phase 1 — Behavior Cloning:**
+  - Input (state vector) = detected entity list (class + position relative to player) + HUD status (from 3.5) + a downsampled version of the raw frame as additional context
+  - Output = predicted input/action, trained to imitate the human recordings
+- **Phase 2 — Autonomous learning (later, design open):** likely reinforcement learning; reward function definition is an open item.
+- **Multi-policy design goal:** architecture should allow multiple distinct agent "personalities" to be trained later on top of the same perception stack (e.g. a "speedrunner" policy, a "score maximizer" policy), without redesigning the perception layer.
+
+---
+
+## 4. Timing & Performance Constraints
+
+- **Hard budget: 66ms for the entire tick** — screen capture → all detection models → high-level logic classifiers → state vector construction → control agent decision → input execution.
+- Detection models should run **in parallel** wherever possible (multi-process/multi-thread or batched on GPU) to maximize how much can happen within the budget.
+- Target hardware for this budget: RTX 3070 Ti (current), assuming sufficient compute headroom; model sizes must be chosen accordingly.
+
+---
+
+## 5. Open Items / Roadmap
+
+- [ ] Research approach for KPI validation via game memory reading (Phase 2 evaluation)
+- [ ] Define reward function(s) for the autonomous learning (RL) phase
+- [ ] Design multi-policy framework for distinct agent behaviors (speedrunner, score-farmer, etc.)
+- [ ] Decide on exact heatmap loss formulation / training details for the anchor-free detector
+- [ ] Long-term: keep model sizes/architectures FPGA-portability-aware (HDMI capture in, USB output) — not an active work item, just a soft constraint to keep in mind
+
+---
+
+## 6. Tech Stack & Conventions
+
+- **Language:** Python (primary), for both recording tooling, labeling backend, training, and inference
+- **Environments:** Ubuntu 24 (recording PC and `jai` training server), `.venv` virtual environments
+- **All project files, code comments, and documentation in English**
+- **Documentation (READMEs) should be kept current as components are built**
+- **Git commits:** every commit message must include the name of the Claude model that authored it
diff --git a/README.md b/README.md
index e69de29..2350df4 100644
--- a/README.md
+++ b/README.md
@@ -0,0 +1,27 @@
+# SpelunkAI
+
+A multi-stage AI system that learns to play *Spelunky Classic HD*: trained perception
+models, deterministic HUD-state classifiers, and a control agent that first imitates
+human play (behavior cloning) and later learns autonomously (reinforcement learning).
+
+Full project spec, architecture, timing constraints, and roadmap live in
+[`CLAUDE.md`](./CLAUDE.md) — that document is the source of truth. This README is just
+a map of the repo.
+
+## Layout
+
+| Directory | Component | Spec section |
+|---|---|---|
+| [`recording/`](./recording) | Recording Tool — captures gameplay video + input log on the recording PC | 3.1 |
+| [`labeling/`](./labeling) | Web-Based Labeling Tool — backend + frontend for bounding-box labeling | 3.2 |
+| [`training/`](./training) | Per-set training pipeline for the anchor-free CNN detectors | 3.3 |
+| [`inference/`](./inference) | Runtime inference loop — capture → detectors → HUD classifiers → state vector, within the 66ms/tick budget | 3.4, 3.5 |
+| [`control/`](./control) | Control Agent — behavior cloning now, RL later | 3.6 |
+
+Each component directory has its own README with more detail and its own `.venv`
+(per `CLAUDE.md` §6), since components run on different machines (recording PC vs.
+`jai`) and have independent dependencies.
+
+## Status
+
+Repo scaffold only — no components are implemented yet.
diff --git a/control/README.md b/control/README.md
new file mode 100644
index 0000000..fe6c206
--- /dev/null
+++ b/control/README.md
@@ -0,0 +1,19 @@
+# Control Agent
+
+Consumes the state vector produced by the inference loop and predicts an action.
+Phase 1: behavior cloning, trained to imitate human input recordings. Phase 2 (later,
+design open): autonomous reinforcement learning. Architecture should allow multiple
+distinct agent "personalities" (e.g. speedrunner, score maximizer) trained later on
+top of the same perception stack.
+
+See CLAUDE.md §3.6 for full requirements.
+
+**Status:** not yet implemented.
+
+## Setup
+
+```
+python -m venv .venv
+source .venv/bin/activate
+pip install -e .
+```
diff --git a/control/pyproject.toml b/control/pyproject.toml
new file mode 100644
index 0000000..cf0f70c
--- /dev/null
+++ b/control/pyproject.toml
@@ -0,0 +1,13 @@
+[project]
+name = "spelunkai-control"
+version = "0.0.0"
+description = "SpelunkAI control agent (behavior cloning now, RL later)"
+requires-python = ">=3.10"
+dependencies = []
+
+[build-system]
+requires = ["setuptools>=68"]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools.packages.find]
+where = ["src"]
diff --git a/control/src/spelunkai_control/__init__.py b/control/src/spelunkai_control/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/inference/README.md b/inference/README.md
new file mode 100644
index 0000000..b1f5df7
--- /dev/null
+++ b/inference/README.md
@@ -0,0 +1,19 @@
+# Inference / Runtime Loop
+
+The live tick loop: screen capture → parallel detection models (per label set) →
+high-level HUD/status logic classifiers (health, gold, bombs, ropes, level — fixed UI
+regions) → state vector construction (entity list + relative positions, HUD status,
+downsampled raw frame). Must fit the entire perception → state-vector pipeline inside
+the 66ms/tick compute budget (~15Hz).
+
+See CLAUDE.md §3.4, §3.5, §4 for full requirements.
+
+**Status:** not yet implemented.
+
+## Setup
+
+```
+python -m venv .venv
+source .venv/bin/activate
+pip install -e .
+```
diff --git a/inference/pyproject.toml b/inference/pyproject.toml
new file mode 100644
index 0000000..eb959f0
--- /dev/null
+++ b/inference/pyproject.toml
@@ -0,0 +1,13 @@
+[project]
+name = "spelunkai-inference"
+version = "0.0.0"
+description = "SpelunkAI runtime inference loop (capture, detectors, HUD classifiers, state vector)"
+requires-python = ">=3.10"
+dependencies = []
+
+[build-system]
+requires = ["setuptools>=68"]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools.packages.find]
+where = ["src"]
diff --git a/inference/src/spelunkai_inference/__init__.py b/inference/src/spelunkai_inference/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/labeling/README.md b/labeling/README.md
new file mode 100644
index 0000000..61c4485
--- /dev/null
+++ b/labeling/README.md
@@ -0,0 +1,16 @@
+# Web-Based Labeling Tool
+
+Bounding-box labeling tool for training frames, with a hierarchical ad-hoc class
+system (Main → Sub, e.g. `Enemy` → `Bat`), multiple independent label sets (Enemy,
+Items, Traps, ...), multi-user support, versioned datasets, and an active-learning
+workflow (seed-label → train → auto-label → review → promote).
+
+See CLAUDE.md §3.2 for full requirements.
+
+**Status:** not yet implemented.
+
+## Structure
+
+- [`backend/`](./backend) — API + data model (sets, hierarchical classes, dataset
+  versions, per-image/per-set label state)
+- [`frontend/`](./frontend) — labeling UI (framework not yet decided)
diff --git a/labeling/backend/pyproject.toml b/labeling/backend/pyproject.toml
new file mode 100644
index 0000000..6f16138
--- /dev/null
+++ b/labeling/backend/pyproject.toml
@@ -0,0 +1,13 @@
+[project]
+name = "spelunkai-labeling-backend"
+version = "0.0.0"
+description = "SpelunkAI Labeling Tool backend: API + data model for bounding-box labels"
+requires-python = ">=3.10"
+dependencies = []
+
+[build-system]
+requires = ["setuptools>=68"]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools.packages.find]
+where = ["src"]
diff --git a/labeling/backend/src/spelunkai_labeling_backend/__init__.py b/labeling/backend/src/spelunkai_labeling_backend/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/labeling/frontend/README.md b/labeling/frontend/README.md
new file mode 100644
index 0000000..f55e3ab
--- /dev/null
+++ b/labeling/frontend/README.md
@@ -0,0 +1,9 @@
+# Labeling Tool — Frontend
+
+UI for the bounding-box labeling tool (image canvas with drawable/editable boxes,
+class picker, dataset/set navigation). Framework not yet chosen — deferred until this
+component is actively built.
+
+See CLAUDE.md §3.2 for full requirements.
+
+**Status:** not yet implemented.
diff --git a/recording/README.md b/recording/README.md
new file mode 100644
index 0000000..2425415
--- /dev/null
+++ b/recording/README.md
@@ -0,0 +1,18 @@
+# Recording Tool
+
+Captures *Spelunky Classic HD* gameplay on the dedicated recording PC (Ubuntu 24
+Desktop, fullscreen 1280x720) as a lossless 30fps MP4, alongside a separate input log
+mapped frame-accurately to the video timeline (frame index + timestamp + key state).
+Output must be easy to slice into individual frames for labeling.
+
+See CLAUDE.md §3.1 for full requirements.
+
+**Status:** not yet implemented.
+
+## Setup
+
+```
+python -m venv .venv
+source .venv/bin/activate
+pip install -e .
+```
diff --git a/recording/pyproject.toml b/recording/pyproject.toml
new file mode 100644
index 0000000..7af6e3a
--- /dev/null
+++ b/recording/pyproject.toml
@@ -0,0 +1,13 @@
+[project]
+name = "spelunkai-recording"
+version = "0.0.0"
+description = "SpelunkAI Recording Tool: gameplay video + input log capture"
+requires-python = ">=3.10"
+dependencies = []
+
+[build-system]
+requires = ["setuptools>=68"]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools.packages.find]
+where = ["src"]
diff --git a/recording/src/spelunkai_recording/__init__.py b/recording/src/spelunkai_recording/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/training/README.md b/training/README.md
new file mode 100644
index 0000000..1603c6a
--- /dev/null
+++ b/training/README.md
@@ -0,0 +1,19 @@
+# Detection Model Training
+
+Per-set training pipeline for the anchor-free, center-heatmap-based (CenterNet-style)
+CNN detectors — one model per label set (Enemy, Items, Traps, ...). Custom, small/
+efficient architectures sized around the overall 66ms/tick inference budget. Includes
+KPI evaluation (precision, recall, mAP-equivalent for the heatmap formulation) against
+held-out labeled data.
+
+See CLAUDE.md §3.3–3.4 for full requirements.
+
+**Status:** not yet implemented.
+
+## Setup
+
+```
+python -m venv .venv
+source .venv/bin/activate
+pip install -e .
+```
diff --git a/training/pyproject.toml b/training/pyproject.toml
new file mode 100644
index 0000000..f2ac441
--- /dev/null
+++ b/training/pyproject.toml
@@ -0,0 +1,13 @@
+[project]
+name = "spelunkai-training"
+version = "0.0.0"
+description = "SpelunkAI detector training pipeline (anchor-free, per-set CNNs)"
+requires-python = ">=3.10"
+dependencies = []
+
+[build-system]
+requires = ["setuptools>=68"]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools.packages.find]
+where = ["src"]
diff --git a/training/src/spelunkai_training/__init__.py b/training/src/spelunkai_training/__init__.py
new file mode 100644
index 0000000..e69de29