Jibun Corp's AI Hub Hits 31 Providers: Adding Replicate and Coze

Background

Building in public: I've been expanding the AI Hub Edge Function in my "Jibun Corp" life management app. Phase 7 adds Replicate (model hosting platform) and Coze (ByteDance's AI agent platform), bringing the total to 31 providers.

What's New

Replicate

• Endpoint: https://openai-compat.replicate.com/v1/chat/completions
• Default model: meta/llama-4-scout-instruct
• Why: Thousands of open-source models, now accessible via a single OpenAI-compat endpoint

Coze (ByteDance)

• Endpoint: https://api.coze.com/v1/chat/completions
• Why: ByteDance's AI agent platform growing fast in Japan and Asia

The 7-Line Pattern

Every new provider is just 7 lines:

replicate: {
  displayName: "Replicate",
  envKey: "REPLICATE_API_TOKEN",
  chatUrl: "https://openai-compat.replicate.com/v1/chat/completions",
  defaultModel: "meta/llama-4-scout-instruct",
  buildBody: OPENAI_COMPAT_BODY,
  parseResponse: OPENAI_COMPAT_PARSE,
},

OPENAI_COMPAT_BODY builds the standard request. OPENAI_COMPAT_PARSE extracts choices[0].message.content. Zero custom code needed for OpenAI-compat providers.

Provider Count Progress

Phase	Count	Added
Phase 1	14	Major LLMs
Phase 5	23	Specialized + infra
Phase 6	29	Meta Llama API, Nebius
Phase 7	31	Replicate, Coze

Takeaway

The OpenAI-compat pattern makes scaling to 31 providers trivial. Each new provider adds ~7 lines, no custom auth code, no custom parsing.

Building in public: https://my-web-app-b67f4.web.app/

FlutterWeb #Supabase #buildinpublic #AI