Hey all! I just published onyx-text-plugin — a drop-in-style alternative to ai-text-plugin that runs on the free FreeAIXYZ gateway. No API key, no account, no quota.

🔗 https://perchance.org/onyx-text-plugin

What it does

Same call style as ai-text-plugin: instruction, startWith, stopSequences, onChunk/onFinish, outputTo, String-like result with .text / .generatedText / .stop() 24 free models across 8 families (general, coding, reasoning, creative, 18+), with model: "auto" capability-based picking Tool calling: calculator, web search, and clock, with an automatic Up-to-10-round agent loop Real token-by-token SSE streaming, reasoning traces (Logos models), structured error codes Special calls: getMetaObject, listModels (live catalog), executeTool, preload Use it in your generator like this:

freeAI = {import:onyx-text-plugin}
let reply = await root.freeAI({
  instruction: "You are Bob, a gruff blacksmith.\n" + chatLog,
  startWith: "Bob:",
  stopSequences: ["\nUser:"],
  model: "onyx-codex",
  tools: "builtin",
  onChunk: (d) => { replyEl.textContent = d.fullTextSoFar; },
});

The generator page itself is full interactive docs + a live playground (model picker, tool toggles, reasoning fold, API reference, error table) — so you can try everything before importing.

Two honest caveats: stop sequences aren't re-included like ai-text-plugin does, and there's deliberately no vision support — I live-tested the gateway's vision slots and they hallucinated instead of seeing images, so images throw a clear error instead of faking it.

Feedback and bug reports welcome! If you build something with it, drop a link — I'd love to see it.

Would it be possible to add an optional custom OpenAI-compatible API configuration to Perchance AI Coder?

This could be useful for people who prefer a different model or want more control over the AI Coder's behavior, especially given some of the recent concerns about the current AI helper/model.

Ideally, the settings could include:

• Model ID — a text input where users can enter the model name/ID.
• Model selection — optionally allow selecting from models returned by the configured API.
• Base URL — for example, an OpenAI-compatible endpoint such as "https://example.com/v1".
• API key — optional input for providers that require authentication.
• Default Perchance model — keep the existing model as the default, so this would be completely optional.

The important part would be supporting OpenAI-compatible APIs, rather than tying the feature to one specific provider. This would let users connect compatible local or third-party endpoints without requiring Perchance to host or provide every model.

For example, the configuration could look something like:

API Base URL: https://example.com/v1 Model ID: my-coding-model API Key: ********

Then AI Coder could send requests using the standard OpenAI-compatible chat/completions interface.

This could also provide a way for users who prefer an older/different coding model to use their own provider while keeping the normal Perchance experience available for everyone else.

Would the Perchance devs consider adding something like this?