What causes "Failed to fetch models for provider ${provider}" in windmill-labs/windmill?

Check console/network for the proxy response body to see the vendor's error (401 vs 404 vs 429)

Failed to fetch models for provider ${provider}

Error message

Failed to fetch models for provider ${provider}

What it means

For all non-Bedrock providers, fetchAvailableModels requests the OpenAI-compatible `models` listing through Windmill's AI proxy, forwarding the provider and resource path (plus anthropic-version for Anthropic). A non-ok response is logged and rethrown as this Error naming the provider. It means the provider rejected the credential/endpoint used by the proxy.

Source

Thrown at frontend/src/lib/components/copilot/lib.ts:270

		})
	}

	// Standard provider handling
	const endpoint = 'models'
	const models = await fetch(
		`${location.origin}${OpenAPI.BASE}/w/${workspace}/ai/proxy/${endpoint}`,
		{
			signal,
			headers: {
				'X-Resource-Path': resourcePath,
				'X-Provider': provider,
				...(provider === 'anthropic' ? { 'anthropic-version': '2023-06-01' } : {})
			}
		}
	)
	if (!models.ok) {
		console.error('Failed to fetch models for provider', provider, models)
		throw new Error(`Failed to fetch models for provider ${provider}`)
	}

	const data = (await readListing(models, maxBytes)) as { data: ModelResponse[] }
	if (data.data.length > 0) {
		const sortFunc = (provider: AIProvider) => (a: string, b: string) => {
			// First prioritize models in defaultModels array
			const defaultModels = AI_PROVIDERS[provider]?.defaultModels || []
			const aInDefault = defaultModels.includes(a)
			const bInDefault = defaultModels.includes(b)

			if (aInDefault && !bInDefault) return -1
			if (!aInDefault && bInDefault) return 1
			return 0
		}
		switch (provider) {
			case 'openai':
				return data.data
					.filter(

View on GitHub (pinned to e474e8803c)

Solutions

Check console/network for the proxy response body to see the vendor's error (401 vs 404 vs 429)
Re-enter/rotate the API key in the provider resource and test it
For customai, verify the resource's base URL exposes an OpenAI-compatible /models endpoint
Fix billing/quota on the vendor account, or retry on transient 5xx

Defensive patterns

Strategy: try-catch

Try / catch

try {
  models = await fetchAvailableModels(res, ws, provider, signal)
} catch (e) {
  notifyUser(`Model listing failed for ${provider}: check the provider resource API key`)
  models = AI_PROVIDERS[provider]?.defaultModels ?? []
}

Prevention

Test the provider API key in the resource editor before chatting
For customai, confirm the base URL serves an OpenAI-compatible /models route
Watch vendor status pages and quota/billing; 429/402 surface here
Cache a known-good model list to degrade gracefully on outages

When it happens

Trigger: GET /w/<workspace>/ai/proxy/models returns non-ok for providers like openai, anthropic, groq, mistral, openrouter, togetherai, deepseek, azure: invalid API key in the provider resource, 401/403 from the vendor, wrong base URL for customai endpoints, quota/billing issues, or vendor outage (5xx).

Common situations: Expired or revoked API keys; customai resource pointing at a base URL without a /models route; Azure resource missing deployment config; rate-limited or unpaid account; typo'd provider resource selection in the chat settings.

Related errors

AI-assisted analysis of windmill-labs/windmill@e474e8803c (2026-09-03). Data as JSON: /api/errors/8dd12a0e8f387453. Report an issue: GitHub.