; =====================================================================
;  Per-environment override for the AI module.
;
;  Copy to ai.production.ini on the production host and fill in real
;  values. The Registry will prefer this file over ai.ini when
;  APPLICATION_ENV=production.
;
;  This file IS gitignored (or should be — add /application/module/ai/
;  settings/ai.production.ini to .gitignore). Never commit production
;  endpoints or model names to a public repo.
; =====================================================================

[AI]
; Point at YOUR Ollama. Self-hosted, behind a reverse proxy, or local —
; whatever you run. Leave the default localhost in ai.ini and override
; here for production.
ollama.base_url     = "https://your-ollama.example.com"
ollama.timeout      = 90

; If you've registered the Nibiru-flavoured model on your Ollama, point
; chat.model at it. Otherwise the fallback covers you.
chat.model          = "nibiru-coder:1.0"
chat.fallback_model = "qwen2.5-coder:14b"

; In production prefer the larger / better embedding model if you've
; pulled it. mxbai-embed-large > nomic-embed-text for quality.
embed.model         = "nomic-embed-text"

; Tighten the ask budget in production — fewer tokens = faster + cheaper.
chat.max_tokens     = 800

; RAG storage in production: switch to database to share a collection
; across multiple app instances behind a load balancer.
;   "json"     — single JSON file per collection (default, in cache/rag/)
;   "database" — uses ai_rag_collection / ai_rag_chunk tables
rag.storage         = "database"

; Optional fallbacks. The framework module ships without these — set
; them only if you want to swap providers temporarily.
; anthropic.api_key   = "sk-ant-..."
; openai.api_key      = "sk-..."