Annoying and alarming things about OpenCode (wren.wtf)

Article summary

The article discusses the author's negative experience with OpenCode, an open-source coding agent, citing issues with its design, security, and performance. The author highlights problems such as prompt cache misses, pruning, and compaction, which lead to inefficiencies and frustration. The article also touches on the author's opinion on local LLMs and their potential impact on codebase stability. The author concludes by advising against using OpenCode due to its numerous issues.

Main themes

• OpenCode issues
• LLM security
• Coding agent design
• Performance optimization
• Local LLMs
• Codebase stability

What commenters say

• OpenCode has a large number of open issues and it seems that nobody is addressing them, leading to a poor user experience.
• The article's criticism of OpenCode is valid, and alternative solutions such as Pi or Codex forks can be used instead.
• Some users rely on OpenCode for its free models and inference endpoints, and switching to alternative solutions may not be feasible due to financial constraints.
• The use of local LLMs can be beneficial, but it also raises concerns about codebase stability and conceptual fidelity.
• OpenCode's LSP integration is a valuable feature, but its effectiveness depends on the quality of the language server and the specific use case.
• The removal of OpenCode from the openrouter.ai leaderboard may be due to the service's own model routing service, Zen.
• Some users have reported improved performance and fewer bugs after switching from OpenCode to alternative solutions like Pi.
• The article's criticism of OpenCode is overly harsh, and the service still has its strengths and loyal user base.