preferences.py

import bpy

from .backend import Backend
from .tools import LlamaMeshModelManager
from .utils import get_available_models


def get_downloaded_models(self, context):
    items = []
    for model in get_available_models():
        items.append((model, model, f"Use {model} for local inference"))

    if not items:
        items.append(("", "No models downloaded", "Download a model first"))
    return items


def reset_backend(self, context):
    backend = Backend.instance()
    backend.reset()

    props = bpy.context.scene.meshgen_props
    props.state = "READY"
    props.history.clear()

    for window in bpy.context.window_manager.windows:
        for area in window.screen.areas:
            if area.type in ["PREFERENCES", "VIEW_3D"]:
                area.tag_redraw()


class MeshGenPreferences(bpy.types.AddonPreferences):
    bl_idname = __package__

    __annotations__ = {
        "backend_type": bpy.props.EnumProperty(
            name="Backend type",
            description="Select the backend type to use for generation",
            items=[
                (
                    "LOCAL",
                    "Local",
                    "Use local integrated llama_cpp_python backend for inference",
                ),
                (
                    "REMOTE",
                    "Remote",
                    "Use a remote API for inference (Ollama, OpenAI, etc.)",
                ),
            ],
            default="REMOTE",
            update=reset_backend,
        ),
        "current_model": bpy.props.EnumProperty(
            name="Current model",
            description="Select model for local inference",
            items=get_downloaded_models,
            update=reset_backend,
        ),
        "download_repo_id": bpy.props.StringProperty(
            name="Repository ID",
            description="Hugging Face repository ID for the model",
            default="bartowski/Meta-Llama-3.1-8B-Instruct-GGUF",
        ),
        "download_filename": bpy.props.StringProperty(
            name="Filename",
            description="Filename of the model to download",
            default="Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf",
        ),
        "llm_provider": bpy.props.EnumProperty(
            name="Provider",
            description="Select the provider to use for remote inference",
            items=[
                (
                    "huggingface",
                    "Hugging Face",
                    "Use the Hugging Face API for inference",
                ),
                ("ollama", "Ollama", "Use an Ollama server for inference"),
                ("anthropic", "Anthropic", "Use the Anthropic API for inference"),
                ("openai", "OpenAI", "Use the OpenAI API for inference"),
            ],
            default="huggingface",
            update=reset_backend,
        ),
        "huggingface_model_id": bpy.props.StringProperty(
            name="Model ID",
            description="ID of the model to use",
            default="meta-llama/Llama-3.3-70B-Instruct",
            update=reset_backend,
        ),
        "huggingface_api_key": bpy.props.StringProperty(
            name="API key",
            description="Hugging Face API key",
            default="",
            subtype="PASSWORD",
            update=reset_backend,
        ),
        "ollama_endpoint": bpy.props.StringProperty(
            name="Endpoint",
            description="Base URL for the Ollama server",
            default="http://localhost:11434",
            update=reset_backend,
        ),
        "ollama_model_name": bpy.props.StringProperty(
            name="Model name",
            description="Name of the model to use",
            default="gemma3",
            update=reset_backend,
        ),
        "ollama_api_key": bpy.props.StringProperty(
            name="API key (optional)",
            description="Ollama API key (optional, used for private models)",
            default="",
            subtype="PASSWORD",
            update=reset_backend,
        ),
        "anthropic_model_id": bpy.props.StringProperty(
            name="Model ID",
            description="ID of the model to use",
            default="claude-3-5-sonnet-latest",
            update=reset_backend,
        ),
        "anthropic_api_key": bpy.props.StringProperty(
            name="API key",
            description="Anthropic API key",
            default="",
            subtype="PASSWORD",
            update=reset_backend,
        ),
        "openai_model_id": bpy.props.StringProperty(
            name="Model ID",
            description="ID of the model to use",
            default="gpt-4o",
            update=reset_backend,
        ),
        "openai_api_key": bpy.props.StringProperty(
            name="API key",
            description="OpenAI API key",
            default="",
            subtype="PASSWORD",
            update=reset_backend,
        ),
        "downloading": bpy.props.BoolProperty(default=False),
        "download_progress": bpy.props.FloatProperty(
            subtype="PERCENTAGE",
            min=0.0,
            max=100.0,
            precision=0,
        ),
        "show_generation_settings": bpy.props.BoolProperty(
            name="Show Generation Settings",
            description="Show or hide generation settings",
            default=False,
        ),
        "temperature": bpy.props.FloatProperty(
            name="Temperature",
            description="Controls randomness in generation (higher = more creative, lower = more predictable)",
            default=0.7,
            min=0.0,
            max=1.0,
        ),
        "context_length": bpy.props.IntProperty(
            name="Context Length",
            description="Controls the maximum number of tokens in the context window",
            default=32768,
            min=1024,
            max=65536,
        ),
        "show_integrations_settings": bpy.props.BoolProperty(
            name="Show Integrations Settings",
            description="Show or hide integrations settings",
            default=False,
        ),
        "enable_hyper3d": bpy.props.BoolProperty(
            name="Enable Hyper3D",
            description="Enable Hyper3D for mesh generation",
            default=False,
        ),
        "hyper3d_api_key": bpy.props.StringProperty(
            name="API key",
            description="Hyper3D API key",
            default="awesomemcp",
            subtype="PASSWORD",
        ),
    }

    def draw(self, context):
        layout = self.layout

        backend_box = layout.box()
        backend_box.label(text="Backend", icon="SETTINGS")

        backend_box.prop(self, "backend_type", expand=True)

        if self.backend_type == "LOCAL":
            info_box = backend_box.box()
            info_box.label(text="Run models directly in Blender.", icon="INFO")
            col = info_box.column(align=True)
            col.label(text="Requires at least 8GB VRAM.")
            col.label(text="Small local models may struggle with agentic behaviors.")

            local_box = layout.box()

            header = local_box.row(align=True)
            header.label(text="Downloaded models", icon="PACKAGE")
            header.operator("meshgen.open_models_folder", text="", icon="FILE_FOLDER")

            models = get_available_models()
            if models:
                local_box.prop(self, "current_model")
            elif not self.downloading:
                local_box.label(text="No models downloaded.", icon="INFO")

                op = local_box.operator(
                    "meshgen.download_model",
                    text="Download Recommended Model",
                    icon="IMPORT",
                )
                op.repo_id = self.download_repo_id
                op.filename = self.download_filename

            if self.downloading:
                row = local_box.row(align=True)
                row.label(text="Downloading...")
                row.prop(self, "download_progress", slider=True, text="")

        else:
            remote_box = layout.box()

            remote_box.prop(self, "llm_provider")
            remote_box.separator()

            if self.llm_provider == "ollama":
                info_box = remote_box.box()
                info_box.label(
                    text="Run models with a local Ollama server.", icon="INFO"
                )
                col = info_box.column(align=True)
                col.label(text="1. Download and install Ollama from ollama.com")
                col.label(text="2. Run `ollama serve` in the terminal")

                remote_box.separator()

                remote_box.prop(self, "ollama_endpoint")
                remote_box.prop(self, "ollama_model_name")
                remote_box.prop(self, "ollama_api_key")

            elif self.llm_provider == "huggingface":
                info_box = remote_box.box()
                info_box.label(
                    text="Run models with the Hugging Face API.", icon="INFO"
                )
                col = info_box.column(align=True)
                col.label(text="1. Create an account on hf.co")
                col.label(text="2. Go to hf.co/settings/tokens and create a new token")
                col.label(text="3. Paste the token into the API key field")

                remote_box.separator()

                remote_box.prop(self, "huggingface_model_id")
                remote_box.prop(self, "huggingface_api_key")

            elif self.llm_provider == "anthropic":
                info_box = remote_box.box()
                info_box.label(text="Run models with the Anthropic API.", icon="INFO")
                col = info_box.column(align=True)
                col.label(text="1. Create an account on console.anthropic.com")
                col.label(
                    text="2. Go to console.anthropic.com/settings/keys and create a key"
                )
                col.label(text="3. Paste the key into the API key field")

                remote_box.separator()

                remote_box.prop(self, "anthropic_model_id")
                remote_box.prop(self, "anthropic_api_key")

            elif self.llm_provider == "openai":
                info_box = remote_box.box()
                info_box.label(text="Run models with the OpenAI API.", icon="INFO")
                col = info_box.column(align=True)
                col.label(text="1. Create an account on platform.openai.com")
                col.label(
                    text="2. Go to platform.openai.com/api-keys and create a new secret key"
                )
                col.label(text="3. Paste the secret key into the API key field")

                remote_box.separator()

                remote_box.prop(self, "openai_model_id")
                remote_box.prop(self, "openai_api_key")

        options_box = layout.box()
        header = options_box.row(align=True)
        header.prop(
            self,
            "show_generation_settings",
            icon="TRIA_DOWN" if self.show_generation_settings else "TRIA_RIGHT",
            icon_only=True,
            emboss=False,
        )
        header.label(text="Generation Settings")

        if self.show_generation_settings:
            options_box.prop(self, "temperature", slider=True)
            options_box.prop(self, "context_length", slider=True)

        plugin_box = layout.box()
        header = plugin_box.row(align=True)
        header.prop(
            self,
            "show_integrations_settings",
            icon="TRIA_DOWN" if self.show_integrations_settings else "TRIA_RIGHT",
            icon_only=True,
            emboss=False,
        )
        header.label(text="Integrations")

        if self.show_integrations_settings:
            llama_mesh_box = plugin_box.box()

            llama_mesh_box.label(text="LLaMA-Mesh", icon="PACKAGE")

            llama_mesh_box.label(
                text="Use LLaMA-Mesh for local mesh generation and understanding.",
            )

            if LlamaMeshModelManager.instance().is_loaded:
                llama_mesh_box.label(text="LLaMA-Mesh is loaded", icon="CHECKBOX_HLT")
                llama_mesh_box.operator(
                    "meshgen.unload_llama_mesh", text="Unload LLaMA-Mesh", icon="X"
                )
            else:
                if self.backend_type == "LOCAL":
                    llama_mesh_box.label(
                        text="Requires 5GB additional VRAM. Not recommended with local backend.",
                        icon="ERROR",
                    )

                llama_mesh_box.operator(
                    "meshgen.load_llama_mesh", text="Load LLaMA-Mesh", icon="IMPORT"
                )

            hyper3d_box = plugin_box.box()
            hyper3d_box.label(text="Hyper3D", icon="PACKAGE")

            hyper3d_box.label(
                text="Use Hyper3D (Rodin) API for mesh generation.",
            )

            hyper3d_box.prop(self, "enable_hyper3d")
            if self.enable_hyper3d:
                hyper3d_box.prop(self, "hyper3d_api_key")


def register():
    bpy.utils.register_class(MeshGenPreferences)


def unregister():
    bpy.utils.unregister_class(MeshGenPreferences)