# Not Diamond Not Diamond is an intelligent [[Model routing]] service aimed at coding agents. It predicts which model should answer each individual input, then sends it there, cutting cost while holding output quality steady. Reported savings sit around 20 to 25%, with accuracy gains of a few percent. ## The architectural difference Most routers are proxies. Your prompt goes to them, they forward it, and they now sit permanently in your data path. Not Diamond works differently: it integrates through an API alongside whichever gateway and harness you already use, so it decides *where* a request should go without becoming the pipe the request travels through. That distinction matters for anyone who has thought carefully about custodians. A routing decision needs far less information than a full proxy hop, and every hop you remove is one fewer company holding your prompts. The company is SOC-2 and ISO 27001 certified on top of that. ## What it means in practice You keep [[OpenRouter]], [[LiteLLM]] or your direct provider contracts. Not Diamond becomes the brain choosing between them rather than a replacement for them. Listed users include Hugging Face, Dropbox, IBM and OpenRouter itself, which tells you the positioning is deliberately complementary. For [[Claude Code]] style workloads, where a single session mixes trivial file reads with genuinely hard reasoning, per-input routing is where the real money hides. A flat "always use the big model" policy is the most expensive default in agentic work. ## References - Official website: https://www.notdiamond.ai/ - Documentation: https://docs.notdiamond.ai/ - Pricing: https://www.notdiamond.ai/pricing - Python client repository: https://github.com/Not-Diamond/notdiamond-python - Model routing reading list: https://github.com/Not-Diamond/awesome-ai-model-routing ## Related - [[Model routing]] - [[AI Gateway]] - [[OpenRouter]] - [[LiteLLM]] - [[Claude Code]] - [[AI Agents]] - [[Ramp Router]]