OpenCode users seek faster code indexing

This looks less like a raw model-speed problem and more like a context-discovery problem. OpenCode already leans on project bootstrapping and search tools, but the post shows the gap between “can chat well” and “can navigate a codebase efficiently.”

• –OpenCode’s `/init` analyzes the project and writes an `AGENTS.md` file, which helps with repo structure and conventions, but it is not the same as a persistent semantic index.
• –The docs point to fuzzy file search, `grep`/`glob`/`list`, and LSP-driven navigation, so the agent’s speed depends heavily on how well those tools are being used.
• –If the model is moving at 30 t/s but still taking 5+ minutes, the bottleneck is probably tool chatter, context inflation, and repo discovery rather than token generation.
• –For local models like Qwen3-Coder-Next, tighter prompts, better project instructions, and LSP-enabled tooling are likely to matter more than waiting for a magical “codebase index.”
• –The thread also hints at a broader truth: agentic coding quality is as much about orchestration and retrieval as it is about the base model.