Fuse
Give your AI agents a budget they can't exceed
Start new thread

Fuse - Give your AI agents a budget they can't exceed

by•
Fuse is a prepaid API wallet for AI agents. Top up a wallet, generate proxy keys with hard spend caps, and point your agent at one URL. If a key's budget runs out, the call gets blocked before it reaches the provider, not billed after. One URL, 9 providers (OpenAI, Anthropic, Groq, xAI, Kimi, DeepSeek, Gemini and more), 50+ models. Works with LangChain, CrewAI, and any OpenAI-compatible SDK. Just swap the base_url. $10 free credit to start.

Add a comment

Replies

Best
Maker
📌
Hi Product Hunt, I'm Lawal, the solo founder of Fuse. I built this because I kept getting scared of my own AI agents. You hand an agent an API key, it hits a loop or a long chain of calls, and you wake up to a bill you never agreed to. The stories are real. One team reportedly ran up a 500 million dollar Claude bill in a single month, and Uber reportedly burned through its entire 2026 Claude Code budget by April. The root problem is that API keys have no ceiling. Your agent can spend infinitely until you happen to notice. Fuse fixes that. You top up a prepaid wallet, generate proxy keys with hard spend caps, and point your agent at one URL. If a key's budget runs out, the call gets blocked before it reaches the provider. Not billed after. Blocked before. A few things I focused on: One URL across 9 providers (OpenAI, Anthropic, Groq, Cohere, xAI, Kimi, DeepSeek, Gemini), 50+ models Hard per-key spend caps, so one runaway agent can't drain everything Your real API keys stay in a vault and never touch your agents Works with LangChain, CrewAI, Haystack, and any OpenAI-compatible SDK by just swapping the base_url The biggest thing I learned building this: the hard part isn't the proxy, it's getting the billing exactly right down to the cent. People won't trust a wallet that might silently overcharge them, so I spent a lot of time making the ledger provably correct. Every new account gets 10 dollars in free credit to try it. I would genuinely love your feedback, especially on this: what would make you trust a proxy like this with your API keys and your budget? getfuse.online

Having a usage dashboard per key with breakdowns by model and provider would be huge, especially if I can see real-time spend so I know which agent is burning through budget before it hits the cap.

Maker

 Love this, it's exactly the direction Fuse is built for. You already get per-key usage broken down by model and provider in the dashboard, every call logged with its cost, so you can see which key (and which agent) is doing the spending.

Real-time spend visibility is the piece I'm actively sharpening. Right now the dashboard refreshes on a short interval rather than truly live, and "catch the burn before it hits the cap" is high on the list, threshold alerts at, say, 80% of a key's budget so you get warned before it blocks.

Genuinely helpful to hear this framed as a priority. Out of curiosity, would you want that as an in-dashboard alert, a webhook, an email, or all three?

The hard spend caps are exactly what I needed after getting burned by surprise bills last month. One thing that would make this even better is a simple per-team or per-project spend dashboard so I can see which agent is actually using what, instead of just seeing a total. Right now I have to guess based on the key names.

Maker
This is the second time today someone's asked for exactly this, so you're clearly onto something real. You can already isolate spend per key (one key per agent/team/project, each with its own cap and full breakdown by model and provider), but you're right that today you're reading it off key names rather than proper team or project grouping. A cleaner per-project view that aggregates keys under a project, so you see spend by project at a glance instead of decoding key labels, is exactly the kind of thing I want to build next. Really appreciate you framing the pain concretely. When you set up keys today, are you already naming them per-team, or would you rather assign them to a project explicitly?

love the hard spend caps blocking calls before they hit the provider, that is the kind of guardrail that makes me actually trust handing my agents a wallet. swapping one base_url across 9 providers is a quietly brilliant move.

Maker
Thank you, that's exactly the reaction I was going for. The whole bet is that people won't hand an agent a wallet unless the guardrail is real, so blocking the call before it reaches the provider (not billing you after) was non-negotiable. And yeah, the one-base_url-across-9-providers thing means you swap a single line and your existing SDK just works. Appreciate you getting it.