Grind a huge one-time job overnight on a free tier's tiny rate limit

The Stack

Steps

1. 1

   Size the batch to the rate limit and the monthly budget

   State how many items you have, the free rate ceiling, your per-item token estimate, and the monthly token budget. CI confirms the job fits the budget and reports how many hours it will take at that rate, so you know it finishes overnight rather than stalling. Running the batch needs your key and is fenced.

   cat > batch.json <<'JSON'
   {
     "provider": "mistral",
     "items": 3000,
     "rate_limit_rpm": 2,
     "est_tokens_per_item": 1200,
     "monthly_token_budget": 1000000000
   }
   JSON
   node -e '
   const fs = require("fs");
   const c = JSON.parse(fs.readFileSync("batch.json", "utf8"));
   function bad(m) { console.error("BAD: " + m); process.exit(1); }
   const items = c.items || 0;
   const rpm = c.rate_limit_rpm || 0;
   const per = c.est_tokens_per_item || 0;
   const budget = c.monthly_token_budget || 0;
   if (items < 1) bad("no items to process");
   if (rpm < 1) bad("rate_limit_rpm must be at least 1");
   if (per < 1) bad("est_tokens_per_item must be set");
   const used = items * per;
   if (budget && used > budget) bad("job needs " + used + " tokens but the monthly budget is " + budget);
   const etaH = (items / (rpm * 60)).toFixed(1);
   console.log("config OK: " + items + " item(s) at " + rpm + " req/min ~" + etaH + "h, using " + used + " of " + budget + " monthly tokens");
   '

2. 2

   Queue it and let it trickle overnight (the model step, not checked by CI)

   Point your paced runner at the backlog with your Mistral key and let it grind at the rate limit while you sleep. Check the output in the morning. The run is fenced.