ruvnet/ruflo · error

Only 'openai' endpoint type is supported in this build

Error message

Only 'openai' endpoint type is supported in this build

What it means

Even when a model has endpoints, this build only implements the "openai" endpoint type. getEndpoint() checks endpoints[0].type and throws on anything else (models.ts:140) — endpoint families from upstream huggingface/chat-ui ("hf_inference", "anthropic", "vertex", "tgi", …) are not compiled into this OpenAI-only build.

Solutions

1. Change the endpoint type to "openai" and point baseURL at an OpenAI-compatible gateway for that provider (most providers expose one)
2. Remove the unsupported endpoint entry so the default OPENAI_BASE_URL endpoint applies
3. Grep MODELS/config for non-openai type values and fix them all at once
4. Add a startup validation that rejects models whose endpoints[0].type !== "openai" so they never reach the picker

Example fix

// before
"endpoints": [{ "type": "hf_inference", "model": "meta-llama/Llama-3.1-8B-Instruct" }]

// after
"endpoints": [{
	"type": "openai",
	"baseURL": "https://router.huggingface.co/v1",
	"completion": "chat_completions"
}]

const SUPPORTED = new Set(["openai"]);
for (const m of modelsRaw) {
	const bad = m.endpoints?.filter((e) => !SUPPORTED.has(e.type)) ?? [];
	if (bad.length) throw new Error(`Unsupported endpoint type(s) ${bad.map((e) => e.type).join(",")} on ${m.id}`);
}

function isOpenAiEndpoint(e: { type: string }): boolean {
	return e.type === "openai";
}

function isServableModel(m: { endpoints?: { type: string }[] }): boolean {
	return (m.endpoints?.length ?? 0) > 0 && m.endpoints!.every(isOpenAiEndpoint);
}

When it happens

Trigger: A MODELS override (or imported model config) carrying "endpoints": [{ "type": "hf_inference" | "anthropic" | "tgi" | ... }]; the model appears in the picker and fails the moment a generation requests its endpoint.

Common situations: Porting model YAML/JSON from huggingface/chat-ui or another fork that supports many provider types; renaming providers after an upstream refactor; copy-pasting an example config from incompatible docs.

Related errors

• No endpoints configured. This build requires…
• Invalid completion type
• No models available to select task model
• OPENAI_BASE_URL not set
• agents must have ≥1 entry

AI-assisted analysis of ruvnet/ruflo@fa13ee4ad6 (2026-08-18). Data as JSON: /api/errors/0f6c089b3e03688b. Report an issue: GitHub.

Appendix: source

Thrown at ruflo/src/ruvocal/src/lib/server/models.ts:140

	...m,
	chatPromptRender: await getChatPromptRender(m),
	id: m.id || m.name,
	displayName: m.displayName || m.name,
	preprompt: m.prepromptUrl ? await fetch(m.prepromptUrl).then((r) => r.text()) : m.preprompt,
	parameters: { ...m.parameters, stop_sequences: m.parameters?.stop },
	unlisted: m.unlisted ?? false,
});

const addEndpoint = (m: Awaited<ReturnType<typeof processModel>>) => ({
	...m,
	getEndpoint: async (): Promise<Endpoint> => {
		if (!m.endpoints || m.endpoints.length === 0) {
			throw new Error("No endpoints configured. This build requires OpenAI-compatible endpoints.");
		}
		// Only support OpenAI-compatible endpoints in this build
		const endpoint = m.endpoints[0];
		if (endpoint.type !== "openai") {
			throw new Error("Only 'openai' endpoint type is supported in this build");
		}
		return await endpoints.openai({ ...endpoint, model: m });
	},
});

type InternalProcessedModel = Awaited<ReturnType<typeof addEndpoint>> & {
	isRouter: boolean;
	hasInferenceAPI: boolean;
};

const inferenceApiIds: string[] = [];

const getModelOverrides = (): ModelOverride[] => {
	const overridesEnv = (Reflect.get(config, "MODELS") as string | undefined) ?? "";

	if (!overridesEnv.trim()) {
		return [];
	}

View on GitHub (pinned to fa13ee4ad6)