DeepClaude – Claude Code agent loop with DeepSeek V4 Pro (github.com)

Article summary

DeepClaude is a project that allows users to use Claude Code's autonomous agent loop with DeepSeek V4 Pro, OpenRouter, or other Anthropic-compatible backends, reducing costs by up to 17 times. The project works by swapping the brain of Claude Code with a cheaper alternative, while keeping the same user experience. This is achieved by setting environment variables that determine which API endpoint to use, allowing users to switch between different backends. The project also provides a proxy that tracks token usage and calculates cost savings.

Main themes

• Autonomous coding agents
• Cost reduction
• Alternative backends
• Claude Code
• DeepSeek V4 Pro
• OpenRouter

What commenters say

• Using DeepSeek V4 Pro with Claude Code can significantly reduce costs without sacrificing performance for routine tasks.
• Some users prefer to use open-source alternatives like OpenCode, which offers more features and customization options.
• Others argue that OpenCode's default behavior of sending prompts to their servers for logging and processing is a concern for privacy and security.
• The effectiveness of different models, such as GLM 5.1 and DeepSeek V4 Pro, varies depending on the specific use case and task complexity.
• Some users find that using a combination of models, such as Opus for planning and a cheaper model for implementation, can be an effective cost-reduction strategy.
• The choice of hardware and infrastructure, such as mini PCs and GPUs, can also impact the performance and cost-effectiveness of autonomous coding agents.
• Managing contexts and sessions with multiple models can be a complex and nuanced problem, requiring careful consideration of factors like token usage and compaction.
• The cost savings of using cheaper models may not always be significant, and other factors like performance and ease of use should also be considered.