opencode-cache-timer

Sidebar countdown showing time until your prompt cache expires.

Cache
4m 23s

Install

Add to ~/.config/opencode/tui.json:

{
  "$schema": "https://opencode.ai/tui.json",
  "plugin": ["opencode-cache-timer"]
}

Restart OpenCode. The Cache panel appears in the sidebar.

While iterating locally, point at the source file instead: "plugin": ["/absolute/path/to/opencode-cache-timer/src/index.tsx"]

Configure

Edit src/providers/registry.ts to set TTLs per provider or per model.

export const DEFAULT_REGISTRY: TTLRegistry = {
  // Provider-level: applies to every model under that provider
  anthropic: 5 * MINUTE,
  openai:    5 * MINUTE,
  deepseek:  2 * HOUR,
  google:    1 * HOUR,
  xai:       5 * MINUTE,

  // Per-model override (wins over provider-level)
  "anthropic:claude-opus-4-6": 1 * HOUR,
  "openai:gpt-5.4":            24 * HOUR,
}

Lookup order: providerID:modelID → providerID → no timer.

If a provider isn't in the registry, the panel shows ◦ idle and no countdown runs. Add an entry to enable it.

Defaults shipped

| Provider | TTL | Notes | |---|---|---| | Anthropic | 5 min | Default ephemeral cache. Override per-model for the 1h tier. | | OpenAI | 5 min | Conservative; actual 5–10 min. Override per-model for the 24h tier. | | DeepSeek | 2 hrs | Conservative; docs say "hours to days". | | Google Gemini | 1 hr | Explicit Context Caching API default. | | xAI Grok | 5 min | No SLA — treated as ephemeral. | | Mistral | — | Undocumented; intentionally omitted. |

Design bias: conservative. A timer that says expired while the cache is still warm is harmless; a timer counting down on an already-dead cache is misleading.

Develop

bun install
bun test