opencode-lcm stretches local context windows

The core idea is practical rather than flashy: preserve full fidelity off-context, keep only what matters in-context, and stop paying for repeated re-summarization with extra model calls.

• –Good fit for local-model workflows where context decay, not raw model quality, is the bottleneck
• –Extractive summaries are a smart tradeoff for searchability and zero extra inference cost, but they will be terse by design
• –The SQLite archive plus 16 recall tools makes this feel more like session memory infrastructure than a simple prompt compressor
• –The biggest limitation is surface area: it only helps if you are already using OpenCode, so adoption depends on that ecosystem
• –Pairing it with context-mode is the right architectural move because it solves both context bloat and output flood