config

type BrainConfig struct {
	// Enabled controls whether the brain is active. Default: false.
	Enabled bool `json:"enabled"`

	// IntelligenceMode controls RAM residency and model quality tier.
	// "optimal" (<800 MB, 8 GB systems), "standard" (~2.2 GB, 16 GB systems),
	// "full" (~4.5 GB, 32 GB+ systems). Leave empty to use legacy auto-scaling.
	IntelligenceMode IntelligenceMode `json:"intelligence_mode,omitempty"`

	// OllamaURL is the base URL of the Ollama server. Default: "http://localhost:11434".
	OllamaURL string `json:"ollama_url,omitempty"`

	// Model is the primary model tag (enrichment fallback when ModelEnrich is unset).
	// Default: "qwen3.5:2b" — base model until fine-tuned SIL models are available.
	Model string `json:"model,omitempty"`

	// FastModel is the model tag for bulk ingestion (fallback when ModelIngest is unset).
	// Default: "qwen3.5:2b" — unified single model until fine-tuned tiers are ready.
	FastModel string `json:"fast_model,omitempty"`

	// ModelIngest is the model for bulk node summarization at index time.
	// Tier 0 (Reflex): simple extraction, no reasoning needed. Default: "qwen3.5:2b".
	// Future: fine-tuned Sentry 0.5B (see PLAN-5-MODELS.md).
	ModelIngest string `json:"model_ingest,omitempty"`

	// ModelGuardian is the model for rule violation explanations.
	// Tier 1 (Sensory): structured plain-English output. Default: "qwen3.5:2b".
	// Future: fine-tuned Critic 1.5B (see PLAN-5-MODELS.md).
	ModelGuardian string `json:"model_guardian,omitempty"`

	// ModelEnrich is the model for architectural enrichment and insight generation.
	// Tier 2 (Specialist): complex analysis across multiple callers/callees. Default: "qwen3.5:2b".
	// Future: fine-tuned Librarian 1.5B (see PLAN-5-MODELS.md).
	ModelEnrich string `json:"model_enrich,omitempty"`

	// ModelOrchestrate is the model for multi-agent conflict resolution.
	// Tier 3 (Architect): deep reasoning about competing scope claims. Default: "qwen3.5:2b".
	// Future: fine-tuned Navigator 2B (see PLAN-5-MODELS.md).
	ModelOrchestrate string `json:"model_orchestrate,omitempty"`

	// ModelArchivist is the model for session memory synthesis.
	// Tier 2 (Specialist): analyzes agent sessions and extracts persistent memories.
	// Default: same as ModelOrchestrate. Future: fine-tuned Archivist 2B.
	ModelArchivist string `json:"model_archivist,omitempty"`

	// Backend selects the LLM backend.
	// "ollama" (default): calls the Ollama HTTP API. Recommended for all users.
	// "local": loads a GGUF file directly via gollama (no Ollama required).
	// Build with -tags llamacpp and CGO_ENABLED=1 for the local backend.
	Backend string `json:"backend,omitempty"`

	// GGUFPath is the path to the fine-tuned GGUF model file.
	// Only used when Backend == "local". If empty, auto-computed as ModelDir/HFFilename.
	// Example: "~/.synapses/models/sil-coder-Q5_K_M.gguf"
	GGUFPath string `json:"gguf_path,omitempty"`

	// ModelDir is the directory where GGUF models are stored.
	// Default: ~/.synapses/models/
	ModelDir string `json:"model_dir,omitempty"`

	// HFRepo is the HuggingFace repository to download the model from.
	// Example: "divish/sil-coder"
	// Used by `brain config download` and `brain serve` (auto-download on first run).
	HFRepo string `json:"hf_repo,omitempty"`

	// HFFilename is the GGUF filename within the HuggingFace repo.
	// Default: "sil-coder-Q5_K_M.gguf"
	HFFilename string `json:"hf_filename,omitempty"`

	// TimeoutMS is the per-request LLM timeout in milliseconds.
	// The HTTP server WriteTimeout is set to 2× this value. Default: 60000 (60s).
	// Must exceed the slowest LLM inference time on your hardware (~25s for 9b CPU).
	TimeoutMS int `json:"timeout_ms,omitempty"`

	// DBPath is the path to the brain's own SQLite database.
	// Default: ~/.synapses/brain.sqlite
	DBPath string `json:"db_path,omitempty"`

	// Port is the HTTP server port for sidecar mode. Default: 11435.
	Port int `json:"port,omitempty"`

	// v0.1.0 feature flags — all default to true when Enabled=true.
	Ingest      bool `json:"ingest"`
	Enrich      bool `json:"enrich"`
	Guardian    bool `json:"guardian"`
	Orchestrate bool `json:"orchestrate"`
	Memorize    bool `json:"memorize"`

	// v0.2.0: Context Packet and SDLC intelligence.
	// ContextBuilder enables BuildContextPacket (default: true when Enabled=true).
	ContextBuilder bool `json:"context_builder"`
	// LearningEnabled enables the decision log and co-occurrence learning (default: true).
	LearningEnabled bool `json:"learning_enabled"`
	// DefaultPhase is the initial SDLC phase stored in brain.sqlite if none is set.
	// Values: "planning" | "development" | "testing" | "review" | "deployment"
	// Default: "development"
	DefaultPhase string `json:"default_phase,omitempty"`
	// DefaultMode is the initial quality mode stored in brain.sqlite if none is set.
	// Values: "quick" | "standard" | "enterprise"
	// Default: "standard"
	DefaultMode string `json:"default_mode,omitempty"`

	// PulseURL is the base URL of the synapses-pulse analytics sidecar.
	// When set, every LLM inference call is reported as a BrainUsageEvent.
	// Leave empty to disable (default). Example: "http://localhost:11437"
	PulseURL string `json:"pulse_url,omitempty"`

	// PulseTimeoutSec is the per-request HTTP timeout for pulse calls.
	// Defaults to 2 when PulseURL is set.
	PulseTimeoutSec int `json:"pulse_timeout_sec,omitempty"`
}

type IntelligenceMode string