bone-agent 1.3.0

package/src/llm/config.py

@@ -0,0 +1,466 @@
+import os
+import platform
+from pathlib import Path
+import yaml
+
+# Provider selection - loaded from config (see after PROVIDER_REGISTRY definition)
+
+def resolve_config_path() -> Path:
+    """Resolve config.yaml path.
+
+    Resolution order:
+    1. BONE_CONFIG_PATH — explicit user config path (set by npm wrapper)
+    2. ~/.bone/config.yaml — user home dotfile (dev and npm installs)
+    """
+    _user_cfg = os.environ.get('BONE_CONFIG_PATH')
+    if _user_cfg:
+        return Path(_user_cfg).resolve()
+    return Path.home() / '.bone' / 'config.yaml'
+
+# Module-level config path (single source of truth)
+CONFIG_PATH = resolve_config_path()
+
+# Environment variable names for API keys (env vars take precedence over config file)
+ENV_API_KEYS = {
+    'ANTHROPIC_API_KEY': os.environ.get('ANTHROPIC_API_KEY'),
+    'OPENAI_API_KEY': os.environ.get('OPENAI_API_KEY'),
+    'GLM_PLAN_API_KEY': os.environ.get('GLM_PLAN_API_KEY'),
+    'GLM_API_KEY': os.environ.get('GLM_API_KEY'),
+    'GEMINI_API_KEY': os.environ.get('GEMINI_API_KEY'),
+    'OPENROUTER_API_KEY': os.environ.get('OPENROUTER_API_KEY'),
+    'KIMI_API_KEY': os.environ.get('KIMI_API_KEY'),
+    'MINIMAX_PLAN_API_KEY': os.environ.get('MINIMAX_PLAN_API_KEY'),
+    'MINIMAX_API_KEY': os.environ.get('MINIMAX_API_KEY'),
+    'BONE_PROXY_API_KEY': os.environ.get('BONE_PROXY_API_KEY'),
+}
+
+# Detect platform for llama.cpp paths
+_IS_WINDOWS = platform.system() == "Windows"
+_IS_LINUX = platform.system() == "Linux"
+
+# Set llama.cpp paths based on platform
+if _IS_WINDOWS:
+    _LLAMA_SERVER_NAME = "llama-server.exe"
+    _LLAMA_BUILD_DIR = "build"
+elif _IS_LINUX:
+    _LLAMA_SERVER_NAME = "llama-server"
+    _LLAMA_BUILD_DIR = "build-linux"
+else:
+    # Fallback for macOS or other platforms
+    _LLAMA_SERVER_NAME = "llama-server"
+    _LLAMA_BUILD_DIR = "build"
+
+def _load_config():
+    """Load config from YAML file, with environment variable overrides for API keys.
+    
+    Environment variables take precedence over values in config.yaml.
+    """
+    config_path = resolve_config_path()
+    if not config_path.exists():
+        return {}
+    try:
+        config = yaml.safe_load(config_path.read_text(encoding="utf-8-sig")) or {}
+    except yaml.YAMLError:
+        config = {}
+    
+    # Override API keys from environment variables (env vars take precedence)
+    for key, env_value in ENV_API_KEYS.items():
+        if env_value:  # Only override if env var is set and non-empty
+            config[key] = env_value
+    
+    return config
+
+
+_CONFIG = _load_config()
+
+# Cache for provider registry (built once at module load)
+_provider_registry_cache = None
+_cached_provider = None
+
+
+def get_model_cost(model_name: str, config: dict | None = None) -> tuple[float, float]:
+    """Get model-specific (cost_in, cost_out) per 1M tokens from MODEL_PRICES.
+
+    Shared utility — used by both the provider registry builder and config_manager.
+    Returns (0.0, 0.0) for unknown models.
+
+    Args:
+        model_name: Model name to look up.
+        config: Optional config dict (defaults to _CONFIG). Pass a fresh config
+                after runtime edits to get up-to-date pricing.
+    """
+    model_prices = (config if config is not None else _CONFIG).get("MODEL_PRICES", {})
+    if model_name in model_prices:
+        mc = model_prices[model_name]
+        return float(mc.get("cost_in", 0.0)), float(mc.get("cost_out", 0.0))
+    return 0.0, 0.0
+
+
+def _model_cost(model_config_key):
+    """Return {cost_in, cost_out} dict for use in provider registry via **spread.
+
+    Reads from module-level _CONFIG (not an injected config). Safe because
+    reload_config() invalidates the provider registry cache, forcing a rebuild.
+    """
+    ci, co = get_model_cost(_CONFIG.get(model_config_key, ""))
+    return {"cost_in": ci, "cost_out": co}
+
+
+def _get_provider_registry():
+    """Build PROVIDER_REGISTRY from current config (cached)."""
+    global _provider_registry_cache
+    if _provider_registry_cache is not None:
+        return _provider_registry_cache
+
+    _provider_registry_cache = {
+        "local": {
+            "type": "local",
+            "api_key": None,
+            "model": _CONFIG.get("LOCAL_MODEL_PATH", ""),
+            "api_model": "model",
+            "api_base": "http://127.0.0.1:8080",
+            "endpoint": "/v1/chat/completions",
+            "error_prefix": "local server",
+            "config_keys": {
+                "LOCAL_MODEL_PATH": "",
+                "LOCAL_SERVER_PATH": str(
+                    Path(__file__).resolve().parents[2] /
+                    f"llama.cpp/{_LLAMA_BUILD_DIR}/bin/{_LLAMA_SERVER_NAME}"
+                ),
+            },
+            "extra": {
+                "host": "127.0.0.1",
+                "port": 8080,
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": True,
+            "allow_temperature": True,
+            "cost_in": 0.0,
+            "cost_out": 0.0
+        },
+        "openrouter": {
+            "type": "api",
+            "api_key": _CONFIG.get("OPENROUTER_API_KEY", ""),
+            "model": _CONFIG.get("OPENROUTER_MODEL", ""),
+            "api_base": _CONFIG.get("OPENROUTER_API_BASE", "https://openrouter.ai/api/v1"),
+            "endpoint": "/chat/completions",
+            "error_prefix": "OpenRouter",
+            "headers_extra": {
+                "HTTP-Referer": "http://localhost:8080",
+                "X-Title": "Chat App"
+            },
+            "config_keys": {
+                "OPENROUTER_API_KEY": "",
+                "OPENROUTER_MODEL": "",
+                "OPENROUTER_API_BASE": "https://openrouter.ai/api/v1",
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": True,
+            "allow_temperature": True,
+            **_model_cost("OPENROUTER_MODEL"),
+        },
+        "glm_plan": {
+            "type": "api",
+            "api_key": _CONFIG.get("GLM_PLAN_API_KEY", ""),
+            "model": _CONFIG.get("GLM_PLAN_MODEL", ""),
+            "api_base": _CONFIG.get("GLM_PLAN_API_BASE", "https://open.bigmodel.cn/api/paas/v4"),
+            "endpoint": "/chat/completions",
+            "error_prefix": "GLM",
+            "config_keys": {
+                "GLM_PLAN_API_KEY": "",
+                "GLM_PLAN_MODEL": "",
+                "GLM_PLAN_API_BASE": "https://open.bigmodel.cn/api/paas/v4",
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": True,
+            "allow_temperature": True,
+            **_model_cost("GLM_PLAN_MODEL"),
+        },
+        "glm": {
+            "type": "api",
+            "api_key": _CONFIG.get("GLM_API_KEY", ""),
+            "model": _CONFIG.get("GLM_MODEL", ""),
+            "api_base": _CONFIG.get("GLM_API_BASE", "https://open.bigmodel.cn/api/paas/v4"),
+            "endpoint": "/chat/completions",
+            "error_prefix": "GLM",
+            "config_keys": {
+                "GLM_API_KEY": "",
+                "GLM_MODEL": "",
+                "GLM_API_BASE": "https://open.bigmodel.cn/api/paas/v4",
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": True,
+            "allow_temperature": True,
+            **_model_cost("GLM_MODEL"),
+        },
+        "openai": {
+            "type": "api",
+            "api_key": _CONFIG.get("OPENAI_API_KEY", ""),
+            "model": _CONFIG.get("OPENAI_MODEL", ""),
+            "api_base": _CONFIG.get("OPENAI_API_BASE", "https://api.openai.com/v1"),
+            "endpoint": "/chat/completions",
+            "error_prefix": "OpenAI",
+            "config_keys": {
+                "OPENAI_API_KEY": "",
+                "OPENAI_MODEL": "",
+                "OPENAI_API_BASE": "https://api.openai.com/v1",
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": False,
+            "allow_temperature": False,
+            **_model_cost("OPENAI_MODEL"),
+        },
+        "gemini": {
+            "type": "api",
+            "api_key": _CONFIG.get("GEMINI_API_KEY", ""),
+            "model": _CONFIG.get("GEMINI_MODEL", ""),
+            "api_base": _CONFIG.get("GEMINI_API_BASE", "https://generativelanguage.googleapis.com/v1beta"),
+            "endpoint": "/chat/completions",
+            "error_prefix": "Gemini",
+            "config_keys": {
+                "GEMINI_API_KEY": "",
+                "GEMINI_MODEL": "",
+                "GEMINI_API_BASE": "https://generativelanguage.googleapis.com/v1beta",
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": True,
+            "allow_temperature": True,
+            **_model_cost("GEMINI_MODEL"),
+        },
+        "minimax_plan": {
+            "type": "api",
+            "api_key": _CONFIG.get("MINIMAX_PLAN_API_KEY", ""),
+            "model": _CONFIG.get("MINIMAX_PLAN_MODEL", ""),
+            "api_base": _CONFIG.get("MINIMAX_PLAN_API_BASE", "https://api.minimax.io/anthropic/v1"),
+            "endpoint": "/messages",
+            "error_prefix": "MiniMax",
+            "config_keys": {
+                "MINIMAX_PLAN_API_KEY": "",
+                "MINIMAX_PLAN_MODEL": "",
+                "MINIMAX_PLAN_API_BASE": "https://api.minimax.io/anthropic/v1",
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": False,
+            "allow_temperature": True,
+            "max_tokens": 4096,
+            **_model_cost("MINIMAX_PLAN_MODEL"),
+        },
+        "minimax": {
+            "type": "api",
+            "api_key": _CONFIG.get("MINIMAX_API_KEY", ""),
+            "model": _CONFIG.get("MINIMAX_MODEL", ""),
+            "api_base": _CONFIG.get("MINIMAX_API_BASE", "https://api.minimax.io/anthropic/v1"),
+            "endpoint": "/messages",
+            "error_prefix": "MiniMax",
+            "config_keys": {
+                "MINIMAX_API_KEY": "",
+                "MINIMAX_MODEL": "",
+                "MINIMAX_API_BASE": "https://api.minimax.io/anthropic/v1",
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": False,
+            "allow_temperature": True,
+            "max_tokens": 4096,
+            **_model_cost("MINIMAX_MODEL"),
+        },
+        "anthropic": {
+            "type": "api",
+            "api_key": _CONFIG.get("ANTHROPIC_API_KEY", ""),
+            "model": _CONFIG.get("ANTHROPIC_MODEL", ""),
+            "api_base": _CONFIG.get("ANTHROPIC_API_BASE", "https://api.anthropic.com/v1"),
+            "endpoint": "/messages",
+            "error_prefix": "Anthropic",
+            "headers_extra": {
+                "anthropic-version": "2023-06-01"
+            },
+            "config_keys": {
+                "ANTHROPIC_API_KEY": "",
+                "ANTHROPIC_MODEL": "",
+                "ANTHROPIC_API_BASE": "https://api.anthropic.com/v1",
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": False,
+            "allow_temperature": True,
+            "max_tokens": 4096,
+            **_model_cost("ANTHROPIC_MODEL"),
+        },
+        "kimi": {
+            "type": "api",
+            "api_key": _CONFIG.get("KIMI_API_KEY", ""),
+            "model": _CONFIG.get("KIMI_MODEL", ""),
+            "api_base": _CONFIG.get("KIMI_API_BASE", "https://api.moonshot.cn/v1"),
+            "endpoint": "/chat/completions",
+            "error_prefix": "Kimi",
+            "config_keys": {
+                "KIMI_API_KEY": "",
+                "KIMI_MODEL": "",
+                "KIMI_API_BASE": "https://api.moonshot.cn/v1",
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": True,
+            "allow_temperature": True,
+            **_model_cost("KIMI_MODEL"),
+        },
+        "bone": {
+            "type": "api",
+            "api_key": _CONFIG.get("BONE_PROXY_API_KEY", ""),
+            "model": _CONFIG.get("BONE_PROXY_MODEL", "openai/gpt-4o-mini"),
+            "api_base": _CONFIG.get("BONE_PROXY_API_BASE", "https://api.vmcode.dev"),
+            "endpoint": "/v1/chat/completions",
+            "error_prefix": "bone-agent Proxy",
+            "config_keys": {
+                "BONE_PROXY_API_KEY": "",
+                "BONE_PROXY_MODEL": "openai/gpt-4o-mini",
+                "BONE_PROXY_API_BASE": "https://api.vmcode.dev",
+            },
+            "default_temperature": 0.1,
+            "default_top_p": 0.9,
+            "allow_top_p": True,
+            "allow_temperature": True,
+            **_model_cost("BONE_PROXY_MODEL"),
+        },
+    }
+    return _provider_registry_cache
+
+
+def _get_provider():
+    """Get the current provider from config (cached)."""
+    global _cached_provider
+    if _cached_provider is not None:
+        return _cached_provider
+
+    last_provider = _CONFIG.get("LAST_PROVIDER")
+    if last_provider and last_provider in _provider_registry_cache:
+        _cached_provider = last_provider
+        return _cached_provider
+    _cached_provider = "glm_plan"
+    return _cached_provider
+
+
+def reload_config():
+    """Reload config from disk and invalidate caches.
+    
+    Reloads both the config.yaml file and environment variables.
+    
+    Note: This is a manual operation - call after config changes.
+    """
+    global _CONFIG, _provider_registry_cache, _cached_provider, PROVIDER_REGISTRY, LLM_PROVIDER, STATUS_BAR_SETTINGS
+    _CONFIG = _load_config()
+    _provider_registry_cache = None
+    _cached_provider = None
+    # Rebuild module-level variables
+    PROVIDER_REGISTRY = _get_provider_registry()
+    LLM_PROVIDER = _get_provider()
+    # Rebuild status bar settings
+    STATUS_BAR_SETTINGS = _build_status_bar_settings()
+
+
+def _build_status_bar_settings():
+    """Build STATUS_BAR_SETTINGS dict from current _CONFIG."""
+    sbs = _CONFIG.get("STATUS_BAR_SETTINGS", {})
+    return {
+        "show_curr_tokens": sbs.get("show_curr_tokens", True),
+        "show_in_tokens": sbs.get("show_in_tokens", True),
+        "show_out_tokens": sbs.get("show_out_tokens", True),
+        "show_total_tokens": sbs.get("show_total_tokens", True),
+        "show_cost": sbs.get("show_cost", True),
+    }
+
+
+def update_status_bar_settings(settings_dict):
+    """Update STATUS_BAR_SETTINGS at runtime and persist to config.
+
+    Args:
+        settings_dict: Dict of settings to update (e.g., {"show_cost": False})
+
+    Returns:
+        Updated STATUS_BAR_SETTINGS dict
+    """
+    global STATUS_BAR_SETTINGS
+    STATUS_BAR_SETTINGS.update(settings_dict)
+    return STATUS_BAR_SETTINGS
+
+
+def get_providers():
+    """Get list of available providers.
+
+    Returns:
+        list: List of provider names from PROVIDER_REGISTRY.
+    """
+    return list(PROVIDER_REGISTRY.keys())
+
+
+# ============================================================================
+# PROVIDER REGISTRY - Centralized provider configuration
+# ============================================================================
+
+# Build provider registry and export as module-level constants (loaded once)
+PROVIDER_REGISTRY = _get_provider_registry()
+LLM_PROVIDER = _get_provider()
+
+
+__all__ = [
+    "CONFIG_PATH",
+    "PROVIDER_REGISTRY",
+    "get_providers",
+    "get_model_cost",
+    "LLM_PROVIDER",
+    "TOOLS_ENABLED",
+    "TOOLS_REQUIRE_CONFIRMATION",
+    "WEB_SEARCH_REQUIRE_CONFIRMATION",
+    "APPROVE_MODES",
+    "APPROVE_MODE_LABELS",
+
+    "get_provider_config",
+    "generate_config_template",
+    "reload_config",
+    "STATUS_BAR_SETTINGS",
+    "update_status_bar_settings",
+]
+
+
+def generate_config_template():
+    """Generate default template for config.json from provider registry."""
+    template = {}
+    for provider, config in PROVIDER_REGISTRY.items():
+        if "config_keys" in config:
+            template.update(config["config_keys"])
+    return template
+
+# Tooling configuration
+TOOLS_ENABLED = True
+TOOLS_REQUIRE_CONFIRMATION = False
+WEB_SEARCH_REQUIRE_CONFIRMATION = False
+
+# Status bar configuration
+STATUS_BAR_SETTINGS = _build_status_bar_settings()
+
+# Tool approval modes
+APPROVE_MODES = ("safe", "accept_edits", "danger")
+CYCLEABLE_APPROVE_MODES = ("safe", "accept_edits")
+APPROVE_MODE_LABELS = {
+    "safe": "Safe",
+    "accept_edits": "Accept Edits",
+    "danger": "Danger",
+}
+
+def get_provider_config(provider: str):
+    """Retrieve the configuration dictionary for a given provider.
+
+    Args:
+        provider (str): Provider name (e.g., 'local', 'openrouter', 'glm', 'openai').
+
+    Returns:
+        dict: Provider config from the PROVIDER_REGISTRY.
+    """
+    return PROVIDER_REGISTRY.get(provider, {})