CodeWhale vs Reasonix: local models or DeepSeek caching?

Compare CodeWhale's Ollama support and approval modes with Reasonix's DeepSeek cache design, batch edits, and Node 22 requirement.

Player not loading? Watch on YouTube

CodeWhale and Reasonix take different approaches to the terminal coding assistant. The speaker presents CodeWhale, formerly DeepSeek TUI, as the more flexible option, with Ollama support for local models and connections to other providers. Reasonix stays tied to the DeepSeek API and focuses on keeping prompt prefixes stable to reduce costs across long sessions. Running its CLI on your computer does not make its main model processing local.

The comparison covers CodeWhale's read-only Plan mode, per-action approvals in Agent mode, and automatic approvals in Yolo mode. It also describes concurrent sub-agents, language-server feedback after edits, workspace sandboxing, and an HTTP/SSE interface for server workflows. The speaker places its rebrand at version 0.8.44.

Reasonix uses TypeScript and Ink, requires Node 22 or higher, and supports macOS, Linux, and Windows. The video also discusses desktop version 0.50.0. Its proposed edits use search-and-replace diffs that users review before applying. Local embedding search can use Ollama, while web search can connect to a self-hosted SearXNG instance.

The architectural discussion explains Reasonix's fixed prefix, append-only conversation log, and temporary reasoning region, alongside tool-call repair and visible escalation to DeepSeek V4 Pro. The speaker cites a single-user case study with a 99.82% cache hit rate; that result is not a guarantee for other workloads.