diff --git a/.gitmodules b/.gitmodules
index b518897..7f12414 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -29,3 +29,7 @@
 [submodule "destrum/third_party/spdlog"]
 	path = destrum/third_party/spdlog
 	url = https://github.com/gabime/spdlog.git
+[submodule "destrum/third_party/tinygltf"]
+	path = destrum/third_party/tinygltf
+	url = https://github.com/syoyo/tinygltf.git
+
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 67a5841..b92082c 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -10,3 +10,4 @@ endif()
 
 add_subdirectory(destrum)
 add_subdirectory(lightkeeper)
+add_subdirectory(TheChef)
\ No newline at end of file
diff --git a/destrum/CMakeLists.txt b/destrum/CMakeLists.txt
index a29070e..e525605 100644
--- a/destrum/CMakeLists.txt
+++ b/destrum/CMakeLists.txt
@@ -96,20 +96,19 @@ target_compile_definitions(destrum
         GLM_ENABLE_EXPERIMENTAL
 )
 
-set(DESTRUM_SHADER_SRC "${CMAKE_CURRENT_LIST_DIR}/assets_src/shaders")
-set(DESTRUM_SHADER_OUT "${CMAKE_CURRENT_LIST_DIR}/assets_runtime/shaders")
+set(ASSETS_SRC_DIR     "${CMAKE_CURRENT_LIST_DIR}/assets_src")
+set(ASSETS_RUNTIME_DIR "${CMAKE_CURRENT_LIST_DIR}/assets_runtime")
 
-include(../cmake/compile_shaders.cmake)
+add_custom_target(cook_assets_clean
+        COMMAND TheChef
+        --input  "${ASSETS_SRC_DIR}"
+        --output "${ASSETS_RUNTIME_DIR}"
+        --clean
+)
 
-compile_glsl_to_spv(destrum "${DESTRUM_SHADER_SRC}" "${DESTRUM_SHADER_OUT}" DESTRUM_SPV)
-add_dependencies(destrum destrum_shaders)
-#
-#include(../cmake/compile_slang_shaders.cmake)
-#
-#compile_slang_shaders(
-#        TARGET compile_shaders
-#        SRC_DIR "${CMAKE_SOURCE_DIR}/destrum/assets_src/shaders"
-#        OUT_DIR "${CMAKE_SOURCE_DIR}/destrum/assets_runtime/shaders"
-#        EXTRA_ARGS -O3
-#        ALL
-#)
+add_custom_target(cook_assets ALL
+        COMMAND TheChef
+        --input  "${ASSETS_SRC_DIR}"
+        --output "${ASSETS_RUNTIME_DIR}"
+        DEPENDS TheChef
+)
diff --git a/destrum/include/destrum/Util/ModelLoader.h b/destrum/include/destrum/Util/ModelLoader.h
new file mode 100644
index 0000000..a8ead79
--- /dev/null
+++ b/destrum/include/destrum/Util/ModelLoader.h
@@ -0,0 +1,423 @@
+#ifndef MODELLOADER_H
+#define MODELLOADER_H
+
+namespace ModelLoader {
+    // CPUMesh loader with tinygltf
+    // - Loads first scene (or default scene), iterates nodes, extracts mesh primitives.
+    // - Handles POSITION/NORMAL/TANGENT/TEXCOORD_0 and indices.
+    // - Computes minPos/maxPos.
+    // - Optionally merges primitives into a single CPUMesh per glTF mesh.
+
+#include <tiny_gltf.h>
+#include <glm/glm.hpp>
+
+#include <cstdint>
+#include <cstring>
+#include <limits>
+#include <optional>
+#include <string>
+#include <unordered_map>
+#include <vector>
+#include <stdexcept>
+#include <iostream>
+
+    struct CPUMesh {
+        std::vector<std::uint32_t> indices;
+
+        struct Vertex {
+            glm::vec3 position;
+            float uv_x{};
+            glm::vec3 normal;
+            float uv_y{};
+            glm::vec4 tangent;
+        };
+
+        std::vector<Vertex> vertices;
+
+        std::string name;
+
+        glm::vec3 minPos;
+        glm::vec3 maxPos;
+    };
+
+    // -------------------- helpers --------------------
+
+    static size_t ComponentSizeInBytes(int componentType) {
+        switch (componentType) {
+            case TINYGLTF_COMPONENT_TYPE_BYTE: return 1;
+            case TINYGLTF_COMPONENT_TYPE_UNSIGNED_BYTE: return 1;
+            case TINYGLTF_COMPONENT_TYPE_SHORT: return 2;
+            case TINYGLTF_COMPONENT_TYPE_UNSIGNED_SHORT: return 2;
+            case TINYGLTF_COMPONENT_TYPE_INT: return 4;
+            case TINYGLTF_COMPONENT_TYPE_UNSIGNED_INT: return 4;
+            case TINYGLTF_COMPONENT_TYPE_FLOAT: return 4;
+            case TINYGLTF_COMPONENT_TYPE_DOUBLE: return 8;
+            default: throw std::runtime_error("Unknown glTF component type");
+        }
+    }
+
+    static int TypeNumComponents(int type) {
+        switch (type) {
+            case TINYGLTF_TYPE_SCALAR: return 1;
+            case TINYGLTF_TYPE_VEC2: return 2;
+            case TINYGLTF_TYPE_VEC3: return 3;
+            case TINYGLTF_TYPE_VEC4: return 4;
+            case TINYGLTF_TYPE_MAT2: return 4;
+            case TINYGLTF_TYPE_MAT3: return 9;
+            case TINYGLTF_TYPE_MAT4: return 16;
+            default: throw std::runtime_error("Unknown glTF type");
+        }
+    }
+
+    // Returns pointer to the first element, plus stride in bytes.
+    static std::pair<const std::uint8_t*, size_t>
+    GetAccessorDataPtrAndStride(const tinygltf::Model& model, const tinygltf::Accessor& accessor) {
+        if (accessor.bufferView < 0) {
+            throw std::runtime_error("Accessor has no bufferView");
+        }
+        const tinygltf::BufferView& bv = model.bufferViews.at(accessor.bufferView);
+        const tinygltf::Buffer& buf = model.buffers.at(bv.buffer);
+
+        const size_t componentSize = ComponentSizeInBytes(accessor.componentType);
+        const int numComps = TypeNumComponents(accessor.type);
+        const size_t packedStride = componentSize * size_t(numComps);
+
+        size_t stride = bv.byteStride ? size_t(bv.byteStride) : packedStride;
+
+        const size_t start =
+                size_t(bv.byteOffset) + size_t(accessor.byteOffset);
+
+        if (start + accessor.count * stride > buf.data.size() + 0ull) {
+            // Note: This check is conservative; stride can be larger than packed.
+            // Still useful to catch obvious corrupt files.
+            // If you want strict validation, check per-element.
+        }
+
+        const std::uint8_t* ptr = buf.data.data() + start;
+        return {ptr, stride};
+    }
+
+    template <typename T>
+    static T ReadAs(const std::uint8_t* p) {
+        T v{};
+        std::memcpy(&v, p, sizeof(T));
+        return v;
+    }
+
+    // Reads VEC3 float accessor element i into glm::vec3.
+    // Allows input component type float only for simplicity (common in glTF).
+    static glm::vec3 ReadVec3Float(const std::uint8_t* base, size_t stride, size_t i) {
+        const std::uint8_t* p = base + i * stride;
+        const float x = ReadAs<float>(p + 0);
+        const float y = ReadAs<float>(p + 4);
+        const float z = ReadAs<float>(p + 8);
+        return glm::vec3{x, y, z};
+    }
+
+    static glm::vec2 ReadVec2Float(const std::uint8_t* base, size_t stride, size_t i) {
+        const std::uint8_t* p = base + i * stride;
+        const float x = ReadAs<float>(p + 0);
+        const float y = ReadAs<float>(p + 4);
+        return glm::vec2{x, y};
+    }
+
+    static glm::vec4 ReadVec4Float(const std::uint8_t* base, size_t stride, size_t i) {
+        const std::uint8_t* p = base + i * stride;
+        const float x = ReadAs<float>(p + 0);
+        const float y = ReadAs<float>(p + 4);
+        const float z = ReadAs<float>(p + 8);
+        const float w = ReadAs<float>(p + 12);
+        return glm::vec4{x, y, z, w};
+    }
+
+    static std::uint32_t ReadIndexAsU32(const tinygltf::Model& model, const tinygltf::Accessor& accessor, size_t i) {
+        auto [base, stride] = GetAccessorDataPtrAndStride(model, accessor);
+        const std::uint8_t* p = base + i * stride;
+
+        switch (accessor.componentType) {
+            case TINYGLTF_COMPONENT_TYPE_UNSIGNED_BYTE:
+                return static_cast<std::uint32_t>(ReadAs<std::uint8_t>(p));
+            case TINYGLTF_COMPONENT_TYPE_UNSIGNED_SHORT:
+                return static_cast<std::uint32_t>(ReadAs<std::uint16_t>(p));
+            case TINYGLTF_COMPONENT_TYPE_UNSIGNED_INT:
+                return static_cast<std::uint32_t>(ReadAs<std::uint32_t>(p));
+            default:
+                throw std::runtime_error("Unsupported index componentType (expected u8/u16/u32)");
+        }
+    }
+
+    static void UpdateBounds(glm::vec3& mn, glm::vec3& mx, const glm::vec3& p) {
+        mn.x = std::min(mn.x, p.x);
+        mn.y = std::min(mn.y, p.y);
+        mn.z = std::min(mn.z, p.z);
+        mx.x = std::max(mx.x, p.x);
+        mx.y = std::max(mx.y, p.y);
+        mx.z = std::max(mx.z, p.z);
+    }
+
+    // -------------------- primitive extraction --------------------
+
+    static CPUMesh LoadPrimitiveIntoCPUMesh(const tinygltf::Model& model,
+                                            const tinygltf::Primitive& prim,
+                                            const std::string& nameForMesh) {
+        CPUMesh out{};
+        out.name = nameForMesh;
+
+        // POSITION is required for our vertex buffer
+        auto itPos = prim.attributes.find("POSITION");
+        if (itPos == prim.attributes.end()) {
+            throw std::runtime_error("Primitive has no POSITION attribute");
+        }
+
+        const tinygltf::Accessor& accPos = model.accessors.at(itPos->second);
+        if (accPos.componentType != TINYGLTF_COMPONENT_TYPE_FLOAT || accPos.type != TINYGLTF_TYPE_VEC3) {
+            throw std::runtime_error("POSITION must be VEC3 float for this loader");
+        }
+
+        const size_t vertexCount = size_t(accPos.count);
+
+        // Optional attributes
+        const tinygltf::Accessor* accNormal = nullptr;
+        const tinygltf::Accessor* accTangent = nullptr;
+        const tinygltf::Accessor* accUV0 = nullptr;
+
+        if (auto it = prim.attributes.find("NORMAL"); it != prim.attributes.end()) {
+            accNormal = &model.accessors.at(it->second);
+            if (accNormal->componentType != TINYGLTF_COMPONENT_TYPE_FLOAT || accNormal->type != TINYGLTF_TYPE_VEC3)
+                accNormal = nullptr; // ignore unsupported
+        }
+        if (auto it = prim.attributes.find("TANGENT"); it != prim.attributes.end()) {
+            accTangent = &model.accessors.at(it->second);
+            if (accTangent->componentType != TINYGLTF_COMPONENT_TYPE_FLOAT || accTangent->type != TINYGLTF_TYPE_VEC4)
+                accTangent = nullptr;
+        }
+        if (auto it = prim.attributes.find("TEXCOORD_0"); it != prim.attributes.end()) {
+            accUV0 = &model.accessors.at(it->second);
+            if (accUV0->componentType != TINYGLTF_COMPONENT_TYPE_FLOAT || accUV0->type != TINYGLTF_TYPE_VEC2)
+                accUV0 = nullptr;
+        }
+
+        // Prepare pointers/strides
+        auto [posBase, posStride] = GetAccessorDataPtrAndStride(model, accPos);
+
+        const std::uint8_t* nrmBase = nullptr;
+        size_t nrmStride = 0;
+        const std::uint8_t* tanBase = nullptr;
+        size_t tanStride = 0;
+        const std::uint8_t* uvBase = nullptr;
+        size_t uvStride = 0;
+
+        if (accNormal) {
+            auto p = GetAccessorDataPtrAndStride(model, *accNormal);
+            nrmBase = p.first;
+            nrmStride = p.second;
+            if (size_t(accNormal->count) != vertexCount) accNormal = nullptr; // mismatch -> ignore
+        }
+        if (accTangent) {
+            auto p = GetAccessorDataPtrAndStride(model, *accTangent);
+            tanBase = p.first;
+            tanStride = p.second;
+            if (size_t(accTangent->count) != vertexCount) accTangent = nullptr;
+        }
+        if (accUV0) {
+            auto p = GetAccessorDataPtrAndStride(model, *accUV0);
+            uvBase = p.first;
+            uvStride = p.second;
+            if (size_t(accUV0->count) != vertexCount) accUV0 = nullptr;
+        }
+
+        // Allocate vertices
+        out.vertices.resize(vertexCount);
+
+        // Bounds init
+        glm::vec3 mn{std::numeric_limits<float>::infinity()};
+        glm::vec3 mx{-std::numeric_limits<float>::infinity()};
+
+        // Fill vertices
+        for (size_t i = 0; i < vertexCount; ++i) {
+            CPUMesh::Vertex v{};
+
+            v.position = ReadVec3Float(posBase, posStride, i);
+            UpdateBounds(mn, mx, v.position);
+
+            if (accNormal) {
+                v.normal = ReadVec3Float(nrmBase, nrmStride, i);
+            } else {
+                v.normal = glm::vec3(0.0f, 1.0f, 0.0f);
+            }
+
+            if (accUV0) {
+                glm::vec2 uv = ReadVec2Float(uvBase, uvStride, i);
+                v.uv_x = uv.x;
+                v.uv_y = uv.y;
+            } else {
+                v.uv_x = 0.0f;
+                v.uv_y = 0.0f;
+            }
+
+            if (accTangent) {
+                v.tangent = ReadVec4Float(tanBase, tanStride, i);
+            } else {
+                v.tangent = glm::vec4(1.0f, 0.0f, 0.0f, 1.0f);
+            }
+
+            out.vertices[i] = v;
+        }
+
+        out.minPos = (vertexCount > 0) ? mn : glm::vec3(0.0f);
+        out.maxPos = (vertexCount > 0) ? mx : glm::vec3(0.0f);
+
+        // Indices (optional; if absent, build linear indices)
+        if (prim.indices >= 0) {
+            const tinygltf::Accessor& accIdx = model.accessors.at(prim.indices);
+            if (accIdx.type != TINYGLTF_TYPE_SCALAR) {
+                throw std::runtime_error("Indices accessor must be SCALAR");
+            }
+
+            out.indices.resize(size_t(accIdx.count));
+            for (size_t i = 0; i < size_t(accIdx.count); ++i) {
+                out.indices[i] = ReadIndexAsU32(model, accIdx, i);
+            }
+        } else {
+            // Non-indexed primitive: make it indexed
+            out.indices.resize(vertexCount);
+            for (size_t i = 0; i < vertexCount; ++i) out.indices[i] = static_cast<std::uint32_t>(i);
+        }
+
+        return out;
+    }
+
+    // Merge "src" into "dst" (offset indices)
+    static void AppendMesh(CPUMesh& dst, const CPUMesh& src) {
+        const std::uint32_t baseVertex = static_cast<std::uint32_t>(dst.vertices.size());
+        dst.vertices.insert(dst.vertices.end(), src.vertices.begin(), src.vertices.end());
+
+        dst.indices.reserve(dst.indices.size() + src.indices.size());
+        for (std::uint32_t idx: src.indices) {
+            dst.indices.push_back(baseVertex + idx);
+        }
+
+        if (dst.vertices.size() == src.vertices.size()) {
+            dst.minPos = src.minPos;
+            dst.maxPos = src.maxPos;
+        } else {
+            // update bounds using src bounds (cheap)
+            dst.minPos = glm::vec3(
+                std::min(dst.minPos.x, src.minPos.x),
+                std::min(dst.minPos.y, src.minPos.y),
+                std::min(dst.minPos.z, src.minPos.z)
+            );
+            dst.maxPos = glm::vec3(
+                std::max(dst.maxPos.x, src.maxPos.x),
+                std::max(dst.maxPos.y, src.maxPos.y),
+                std::max(dst.maxPos.z, src.maxPos.z)
+            );
+        }
+    }
+
+    // -------------------- public API --------------------
+
+    // Option A: return one CPUMesh per *primitive* encountered in the scene
+    static std::vector<CPUMesh> LoadGLTF_CPUMeshes_PerPrimitive(const std::string& path) {
+        tinygltf::TinyGLTF loader;
+        tinygltf::Model model;
+        std::string err, warn;
+
+        bool ok = false;
+        const bool isGLB = (path.size() >= 4 && path.substr(path.size() - 4) == ".glb");
+        if (isGLB) ok = loader.LoadBinaryFromFile(&model, &err, &warn, path);
+        else ok = loader.LoadASCIIFromFile(&model, &err, &warn, path);
+
+        if (!warn.empty()) std::cerr << "tinygltf warn: " << warn << "\n";
+        if (!ok) throw std::runtime_error("Failed to load glTF: " + err);
+
+        int sceneIndex = model.defaultScene >= 0 ? model.defaultScene : 0;
+        if (sceneIndex < 0 || sceneIndex >= int(model.scenes.size())) {
+            throw std::runtime_error("glTF has no valid scene");
+        }
+
+        std::vector<CPUMesh> result;
+        const tinygltf::Scene& scene = model.scenes.at(sceneIndex);
+
+        // Traverse nodes and pull mesh primitives (ignoring node transforms in this basic example)
+        std::vector<int> stack(scene.nodes.begin(), scene.nodes.end());
+        while (!stack.empty()) {
+            int nodeIdx = stack.back();
+            stack.pop_back();
+            const tinygltf::Node& node = model.nodes.at(nodeIdx);
+
+            for (int child: node.children) stack.push_back(child);
+
+            if (node.mesh < 0) continue;
+            const tinygltf::Mesh& mesh = model.meshes.at(node.mesh);
+
+            for (size_t p = 0; p < mesh.primitives.size(); ++p) {
+                const tinygltf::Primitive& prim = mesh.primitives[p];
+                CPUMesh cpu = LoadPrimitiveIntoCPUMesh(model, prim, mesh.name.empty() ? ("mesh_" + std::to_string(node.mesh)) : mesh.name);
+                // Optionally make name unique per primitive
+                cpu.name += "_prim" + std::to_string(p);
+                result.push_back(std::move(cpu));
+            }
+        }
+
+        return result;
+    }
+
+    // Option B: return one CPUMesh per *glTF mesh*, merging all primitives of that mesh into one CPUMesh
+    static std::vector<CPUMesh> LoadGLTF_CPUMeshes_MergedPerMesh(const std::string& path) {
+        tinygltf::TinyGLTF loader;
+        tinygltf::Model model;
+        std::string err, warn;
+
+        bool ok = false;
+        const bool isGLB = (path.size() >= 4 && path.substr(path.size() - 4) == ".glb");
+        if (isGLB) ok = loader.LoadBinaryFromFile(&model, &err, &warn, path);
+        else ok = loader.LoadASCIIFromFile(&model, &err, &warn, path);
+
+        if (!warn.empty()) std::cerr << "tinygltf warn: " << warn << "\n";
+        if (!ok) throw std::runtime_error("Failed to load glTF: " + err);
+
+        int sceneIndex = model.defaultScene >= 0 ? model.defaultScene : 0;
+        if (sceneIndex < 0 || sceneIndex >= int(model.scenes.size())) {
+            throw std::runtime_error("glTF has no valid scene");
+        }
+
+        std::vector<CPUMesh> result;
+
+        const tinygltf::Scene& scene = model.scenes.at(sceneIndex);
+        std::vector<int> stack(scene.nodes.begin(), scene.nodes.end());
+
+        while (!stack.empty()) {
+            int nodeIdx = stack.back();
+            stack.pop_back();
+            const tinygltf::Node& node = model.nodes.at(nodeIdx);
+
+            for (int child: node.children) stack.push_back(child);
+
+            if (node.mesh < 0) continue;
+            const tinygltf::Mesh& mesh = model.meshes.at(node.mesh);
+
+            CPUMesh merged{};
+            merged.name = mesh.name.empty() ? ("mesh_" + std::to_string(node.mesh)) : mesh.name;
+            merged.minPos = glm::vec3(std::numeric_limits<float>::infinity());
+            merged.maxPos = glm::vec3(-std::numeric_limits<float>::infinity());
+
+            bool any = false;
+            for (const tinygltf::Primitive& prim: mesh.primitives) {
+                CPUMesh part = LoadPrimitiveIntoCPUMesh(model, prim, merged.name);
+                if (!any) {
+                    merged = part;
+                    any = true;
+                } else {
+                    AppendMesh(merged, part);
+                }
+            }
+
+            if (any) result.push_back(std::move(merged));
+        }
+
+        return result;
+    }
+}
+
+#endif //MODELLOADER_H
diff --git a/destrum/src/App.cpp b/destrum/src/App.cpp
index 267c3aa..563d2f6 100644
--- a/destrum/src/App.cpp
+++ b/destrum/src/App.cpp
@@ -146,6 +146,7 @@ void App::run() {
                 std::chrono::duration<float>(dt)
             );
             std::this_thread::sleep_until(targetEnd);
+
         }
     }
     gfxDevice.waitIdle();
diff --git a/destrum/third_party/CMakeLists.txt b/destrum/third_party/CMakeLists.txt
index 5d76599..2fc3045 100644
--- a/destrum/third_party/CMakeLists.txt
+++ b/destrum/third_party/CMakeLists.txt
@@ -51,3 +51,5 @@ option(JSON_Install         "Install CMake targets during install step." OFF)
 add_subdirectory(json)
 
 add_subdirectory(spdlog)
+
+add_subdirectory(tinygltf)
\ No newline at end of file
diff --git a/destrum/third_party/tinygltf b/destrum/third_party/tinygltf
new file mode 160000
index 0000000..81bd50c
--- /dev/null
+++ b/destrum/third_party/tinygltf
@@ -0,0 +1 @@
+Subproject commit 81bd50c1062fdb956e878efa2a9234b2b9ec91ec
diff --git a/lightkeeper/src/Lightkeeper.cpp b/lightkeeper/src/Lightkeeper.cpp
index ab0c8fc..dd52110 100644
--- a/lightkeeper/src/Lightkeeper.cpp
+++ b/lightkeeper/src/Lightkeeper.cpp
@@ -51,7 +51,7 @@ void LightKeeper::customInit() {
     meshComp->SetMeshID(testMeshID);
     meshComp->SetMaterialID(testMaterialID);
 
-    testCube->AddComponent<Rotator>(10, 5);
+    // testCube->AddComponent<Rotator>(10, 5);
 
 
     scene.Add(testCube);