Open AI Provider Settings — Custom API Endpoint + Model Name

I'd like to raise what I think is one of the most impactful feature improvements you could make to the AI section of eDirectory: opening up the AI provider configuration to allow a custom API base URL and model name, rather than locking users to just OpenAI.

The OpenAI API standard is to AI what S3 is to object storage

The best way to understand this is by analogy. Years ago, Amazon S3 defined an API standard for object storage. Today, dozens of providers — Cloudflare R2, Backblaze B2, Wasabi, MinIO, and many others — all implement that same S3-compatible API. The result is that any application built to talk to S3 can switch providers by simply changing an endpoint URL and credentials. No code changes. No re-integration work.

The OpenAI /v1/chat/completions API has become exactly the same thing for AI inference. It is now the de facto industry standard protocol, and virtually every major AI provider in the world has adopted it as a drop-in compatible interface.

Western providers

• OpenAI — api.openai.com/v1

• Google Gemini — generativelanguage.googleapis.com/v1beta/openai

• xAI / Grok — api.x.ai/v1

• Mistral AI — api.mistral.ai/v1

• Groq — api.groq.com/openai/v1

• Together AI — api.together.xyz/v1

• Fireworks AI — api.fireworks.ai/inference/v1

• Perplexity — api.perplexity.ai

• OpenRouter — openrouter.ai/api/v1 (routes to 100+ models via one key)

• Ollama (self-hosted) — localhost:11434/v1

• LM Studio (self-hosted) — localhost:1234/v1

Chinese providers (all OpenAI-compatible)

• DeepSeek — api.deepseek.com/v1

• Z.ai / GLM (Zhipu AI) — api.z.ai/api/openai/v1

• Qwen / Alibaba — dashscope.aliyuncs.com/compatible-mode/v1

• MiniMax — api.minimax.chat/v1

• Moonshot / Kimi — api.moonshot.ai/v1

• ByteDance / Doubao — ark.cn-beijing.volces.com/api/v3

• Baidu / ERNIE — qianfan.baidubce.com/v2

• StepFun — api.stepfun.com/v1

• Xiaomi / MiMo — via OpenRouter

• Tencent Hunyuan — api.hunyuan.cloud.tencent.com/v1

They all speak the same protocol. If eDirectory simply exposed a custom API base URL and model name field, every single one of these becomes instantly usable — with zero additional development work on your end. One implementation unlocks the entire ecosystem, now and for every provider that emerges in the future. Just like S3 compatibility did for storage.

The cost case

Right now, users are locked to OpenAI pricing. Here is what that actually costs compared to what is available across the full ecosystem (prices per 1 million tokens, USD):

OpenAI — GPT-5.5
Input: $5.00 / Output: $30.00

Anthropic — Claude Opus 4.8
Input: $5.00 / Output: $25.00

(Not natively OpenAI-compatible — accessible via OpenRouter only)

Google Gemini — Gemini 3.1 Pro
Input: $2.00 / Output: $12.00

Mistral — Mistral Large
Input: $0.80 / Output: $2.40

xAI / Grok — Grok 4.1
Input: $0.20 / Output: $0.50

MiniMax — M2.5
Input: $0.12 / Output: $0.48

Moonshot / Kimi — K2.6
Input: $0.10 / Output: $0.40

ByteDance / Doubao — Seed 2.0 Lite
Input: $0.10 / Output: $0.25

Qwen — Qwen3-Coder
Input: $0.03 / Output: $0.11

DeepSeek — V4 Flash
Input: $0.056 / Output: $0.112

Z.ai / GLM — GLM-4.7-Flash
Input: FREE / Output: FREE

Ollama (self-hosted) — Llama 4 / Qwen3
Input: FREE / Output: FREE

That's a 90× cost difference between OpenAI GPT-5.5 and DeepSeek V4 Flash, with free options available for users already running local models or willing to use Z.ai's free Flash tier. For a directory platform calling AI across thousands of listings, search queries, and chatbot interactions, that saving is enormous and compounds fast at scale.

What I am asking for

It doesn't need to be complicated. A simple "Custom / OpenAI-compatible" option in the AI provider dropdown, with three fields:

1. API Base URL

2. Model name

3. API Key

That's it. No need to hard-code or maintain a growing list of every provider — just give us the open fields and we'll configure it ourselves. The S3 analogy holds here too: you don't need to know about every storage provider to support them, you just need to implement the standard.

This would bring eDirectory in line with how every other modern AI-integrated platform handles provider configuration today, and it future-proofs the platform as the AI landscape continues to evolve.

Would love to hear if this is on the roadmap.

Thanks,
Aaron