opencode-plugboard

Appearance

Limits, budgets and caching ​

The problem. A runaway loop in one app should not use up your quota or starve the others. A poller that sends the same prompt every minute should not pay for it every minute.

Cap concurrency and rate ​

limits caps how many model calls run at once, and how many an app may start per minute. Nothing queues. A call over a cap gets a 429 at once, so the provider never sees it.

jsonc
"limits": {
  "maxConcurrent": 8,
  "perApp": { "maxConcurrent": 2, "requestsPerMinute": 30 },
  "apps": { "coding": { "maxConcurrent": 4 } }
}

coding gets four calls at once and still 30 per minute, because apps.<name> overrides perApp field by field. A refused call carries Retry-After:

HTTP/1.1 429 Too Many Requests
retry-after: 12

{"error": {"message": "Too many requests per minute", "code": "rate_limit_exceeded", "type": "rate_limit_error", "param": null}}

A full concurrency cap answers too_many_requests with retry-after: 1. The code is on the OpenAI and Gemini routes. The Anthropic /v1/messages error has a type and a message only. The counts live in the server process, so a restart clears them.

Cap spend ​

budgets sets ceilings per app for a UTC day, week or month: requests, tokens or cost. It reads the usage records, so usage must stay on.

jsonc
"budgets": {
  "perApp": { "requests": { "day": 1000 }, "tokens": { "month": 20000000 } },
  "apps": { "coding": { "costUsd": { "month": 50 } } }
}

An app over a ceiling gets 429 budget_exceeded until the window resets. The server logs budget.warning at 80%. The check runs before a call starts, so the call that crosses a ceiling is still served. Budgets guard against runaway use. They are not an exact meter. Cost is an estimate at list prices.

Cap the size of one request ​

  • maxTools (default 64) caps the tools one request may carry. maxChoices (default 1) caps how many answers one request may ask for with n, because each is a full model call.
  • limits.maxBodyBytes caps the size of one request body. The default, 32 MiB, is also the most it can be.

Replay repeated answers ​

Pollers and bots often send the same prompt again and again. With responseCache, an identical request from the same app gets the stored answer. No session is created and no provider call is made.

jsonc
"responseCache": { "ttlSeconds": 300, "apps": ["discord-bot"] }

Every answer says whether it came from the cache:

x-plugboard-cache: hit

Only normal text answers are kept. Answers with tool calls, errors, structured output and n above 1 are not. With temperature above 0, a hit repeats the first answer, so limit the cache to apps that want that. The cache is in memory and a restart empties it. A hit takes no concurrency slot and spends no budget.

Keep the provider's prompt cache warm ​

promptCache with affinity: "anchor" gives every request of an app the same cache key. It helps ChatGPT subscription models. Anthropic and Copilot GPT already cache without it.

jsonc
"promptCache": { "affinity": "anchor", "scope": "app" }

All in the configuration reference:

Next: Tools, structured output and images.