AI Model Providers
Live model calls
Section titled “Live model calls”When live AI is enabled, calls route through the local Claude Code CLI by default (the operator’s own subscription-shaped access, no per-token metering to this platform). API backends exist as fallbacks:
- Anthropic API — per-token list pricing.
- DeepSeek API — the cheapest API fallback.
Model ids live in one configuration file, not scattered through the code.
Embeddings
Section titled “Embeddings”Semantic matching uses OpenAI’s text-embedding-3-small. When no embedding key is
configured, the platform degrades to zero vectors: non-semantic ranking
still works, and nothing crashes. Components detect the zero vector and
behave accordingly.
Guardrails apply to every backend
Section titled “Guardrails apply to every backend”Whichever backend serves a live call, the same spend cap, quotas, and kill switches apply. Mock mode costs nothing and is the default.