ollama - 💡(How to fix) Fix ollama launch claude-desktop [2 comments, 3 participants]

What is the issue?

Environment:

Ollama: 0.23.1 Claude Desktop: 1.6259.1 macOS: arm64 (Apple Silicon) Mac: MacBook Air

Description: ollama launch claude-desktop successfully puts Claude Desktop into 3P gateway mode and writes a valid config to ~/Library/Application Support/Claude-3p/configLibrary/, but Claude Desktop rejects all Ollama Cloud models during discovery. Steps to reproduce:

export OLLAMA_API_KEY=<valid_key> ollama launch claude-desktop Claude Desktop opens in Cowork 3P · Gateway mode Model dropdown shows only "Legacy Model" with no selectable models Sending a message fails with: There's an issue with the selected model (claude-sonnet-4-6). It may not exist or you may not have access to it.

Diagnostic details: The gateway config is correctly written: json{ "disableDeploymentModeChooser": true, "inferenceGatewayApiKey": "<redacted>", "inferenceGatewayAuthScheme": "bearer", "inferenceGatewayBaseUrl": "https://ollama.com", "inferenceProvider": "gateway" } The API key works — curl to /v1/models returns 39 models in valid OpenAI-compatible format: json{"object":"list","data":[{"id":"kimi-k2.6","object":"model","created":1774915200,"owned_by":"ollama"},{"id":"deepseek-v4-pro","object":"model","created":1776988800,"owned_by":"ollama"},...]} curl to /api/tags also returns all 39 models successfully. Key log line from ~/Library/Logs/Claude-3p/main.log: [warn] [custom-3p] Gateway /v1/models returned 0 usable models { rawCount: 39 } [info] [custom-3p] Model discovery: not supported by provider; picker = 0 (empty) Claude Desktop fetches all 39 models but filters every single one out as "not usable." It then falls back to claude-sonnet-4-6 which doesn't exist on Ollama Cloud. What I tried (none worked):

Restore and relaunch with fresh API key Full quit and restart of Claude Desktop Manually adding inferenceGatewayModels array to the config — ignored by Claude Desktop Multiple API keys — all return models via curl but fail in Claude Desktop

Additional context: This was working previously. It broke after a recent update (Claude Desktop 1.6259.1 or Ollama 0.23.1). The /v1/models response appears to be standard OpenAI-compatible format, so Claude Desktop's model validator may now require additional fields (e.g., capabilities, context_length) that Ollama Cloud doesn't provide.

Relevant log output

OS

No response

GPU

No response

CPU

No response

Ollama version

No response