[Feature Request] Add DemonRoute as a named provider (OpenAI-compatible, uncensored and roleplay fine-tunes) #1663

Disclosure up front: I'm with DemonRoute, so this is a request to add our own service. DemonRoute is built by NFT Demon Holdings LLC, one of the largest consulting firms in blockchain, AI, and quantum computing, serving celebrities and Fortune 500 companies.

What I'd like to propose

A named DemonRoute provider in Risuai, built the same way the NanoGPT integration from #1343 was: endpoint constants in src/ts/model/providers/, models fetched dynamically from GET https://api.demonroute.com/v1/models (the OpenRouter/NanoGPT pattern, so nothing is registered statically in LLMModels), and requests sent through the existing OpenAI-format request path. No new request format is needed.

Users can already reach DemonRoute today through the Custom Models settings page under Advanced, using URL https://api.demonroute.com/v1/chat/completions, the OpenAI format, and their key, or the OpenAI Reverse Proxy option. A named provider would just save the URL lookup and give them a searchable model picker like the NanoGPT and OpenRouter ones. The API is plain OpenAI chat completions with a bearer key, so it's a base URL and key swap.

Why it's useful for Risuai users

Risuai already has OpenRouter, NanoGPT, and the direct OpenAI, Claude, and Gemini paths, so here's what DemonRoute adds that those don't. All of it is from demonroute.com:

• Uncensored and roleplay fine-tunes are first-class, not curated out. 556 unique open-weight models (1,127 endpoints) on one key, served by Verified Hosts that DemonRoute holds direct agreements with. Frontier providers refuse or soften a lot of normal character and fiction content, and aggregators often don't carry community fine-tunes at all. A few current examples from the catalog (context window, then price per 1M input / output tokens):
  • dphn/Dolphin-Mistral-24B-Venice-Edition: 131,072 tokens, $2.80 / $4.64
  • huihui-ai/Huihui-gemma-4-31B-it-abliterated-v2: 262,144 tokens, $0.48 / $1.44
  • Ateron/Gemma-4-Writers-31B-V2: 32,768 tokens, $1.92 / $5.76
  • openai/gpt-oss-120b: 131,072 tokens, $0.60 / $2.40
• Long contexts for long chats. Several models carry 131k to 262k token windows, which helps lorebooks, long chat histories, and HypaMemory/SupaMemory summaries without paying frontier prices.
• Privacy. Client IPs aren't persisted on the API, prompts are proxied to the serving host and not retained, and usage rows store token counts and cost rather than text. There's no training, fine-tuning, or evals on user prompts, and no data is sold. Details: https://demonroute.com/privacy
• An anonymous account option. Username and password with crypto deposits, no email, so users who don't want their chats tied to an identity have a real option.
• Prepaid with a hard stop at $0. No overdraft and no surprise invoice, so a long session can only spend what's loaded.
• Stable model IDs. Delisted models get a 30-day notice, then the old slug routes to a successor for 90 days, so a model saved on a character or preset doesn't break overnight.

Model IDs are Hugging Face-style family/Name strings with no prefixes, so they can be shown and grouped the same way as OpenRouter's.

Honest caveats: DemonRoute is text-only (no image generation, TTS, or vision endpoints), and quality varies across community fine-tunes, so I'd suggest a mainstream default such as openai/gpt-oss-120b.

Next step

If you're open to it, I'll prepare a PR following the NanoGPT pattern, fill in the PR checklist, test it on the web, local, and node-hosted builds. Happy to adjust the scope, or close this if you'd rather keep providers to the current set. Contact: jeremy@nftdemon.com. Integration guide: https://demonroute.com/integrations