AI
Ollama brings alternative models to Claude Desktop after first attempt failed
The open-weight model runner has relaunched Claude Desktop integration with a new architecture that lets developers run Qwen, DeepSeek, and other open models through Anthropic's app.
The New Stack
Ollama, a platform for running open-weight AI models, has successfully reintroduced integration with Claude Desktop, enabling users to access models from its library—including locally-hosted alternatives—without leaving Anthropic's application. The gateway approach means developers can now leverage models like Qwen, DeepSeek, Kimi, and GLM as alternatives to Anthropic's first-party offerings while maintaining the Claude Desktop experience.
A second attempt after earlier setback#
Ollama's journey toward Claude Desktop compatibility began in January, when the company added support for Anthropic's Messages API, allowing Claude Code to connect to Ollama-hosted models. By April, Anthropic itself opened the door by introducing a developer setting in Claude Desktop that accepted third-party inference gateways.
Ollama moved quickly to exploit this opening with a claude-desktop launch command that configured the app to use Ollama as its backend. The approach appeared to work—until a Claude Desktop update in the following days broke the setup. Anthropic's gateway began rejecting non-Anthropic model IDs, effectively limiting third-party access to Anthropic models only. Ollama withdrew the feature rather than offer a broken product.
Redesigned architecture overcomes earlier limitations#
On August 21, Ollama introduced a new technical approach in version 0.33.0 that sidesteps the previous constraints: a dedicated local proxy specifically for Claude Desktop. Rather than relying on Claude Desktop's own gateway, this proxy layer sits between Ollama and the desktop app, translating requests in a way that preserves access to all available models.
The setup process is designed for simplicity. Users open the Ollama app, select Claude, and toggle the integration on—Ollama then automatically configures Claude Desktop's third-party gateway settings. The Mac menu includes a convenient "Use Ollama models" toggle for quick switching, and users can map individual Claude model slots (Opus 5, Sonnet 5, etc.) to any available Ollama model. Disabling the integration restores the prior Claude configuration.
Why choose alternatives over Claude?#
The launch prompted community questions about the practical rationale for running different models through Claude's interface rather than simply using Claude itself. Ollama highlighted three key drivers: , , and , alongside the ability to deploy models fine-tuned on proprietary company data. Users can select between locally-running models and those served through Ollama Cloud, giving developers flexibility around where computation happens.
