CostLatch is a streaming proxy that blocks spend before tokens burn. Daily and monthly hard caps, per-run budgets, and loop detection that stops the same prompt hammering your API key overnight.
Also set native OpenAI / Anthropic / Cursor caps. CostLatch is a second latch — not a replacement for theirs.
We estimate cost, atomically reserve budget, and refuse the call with zero upstream tokens if you're over.
Identical prompts N times in M minutes get a hard block and an alert — before the overnight melt.
By default we store spend aggregates only. Your prompts are not written to our logs.
Self-serve via Stripe. Cancel anytime.
We\'re finishing hard-budget enforcement and loop detection. Leave your email and we\'ll invite you when CostLatch opens.