Launching today
Silvanexum
Hard budgets that stop runaway AI agent spend
2 followers
Hard budgets that stop runaway AI agent spend
2 followers
AI agents loop, retry and resend growing context, and the bill shows up weeks later. Silvanexum meters every LLM call across OpenAI, Anthropic and Gemini (streaming included), ties cost to the customer, agent and trace behind it, and blocks calls before they break a budget. Plug in by swapping one base URL, or use SDK-only mode so prompts and keys never leave your servers. Python and TypeScript SDKs, an MCP server and an NVIDIA NeMo Relay plugin.
Free Options
Launch Team
AWS Startup AdvisorPersonalized AWS guidance that meets you where youāre already building
Promoted
Hey Product Hunt š I'm Rana, the solo engineer behind Silvanexum.
I built it after watching AI agents do something normal APIs never do: decide on their own how much to spend. A retry loop, a context that keeps growing, a tool that returns junk, and a task that should cost cents costs dollars. Provider dashboards only tell you per API key, after the fact.
Silvanexum sits in front of (or beside) your LLM calls and:
ā¢ meters every call across OpenAI, Anthropic and Gemini, streaming included
ā¢ attributes cost to the customer, agent and trace that caused it
ā¢ blocks a call before it runs when a budget is spent
Two ways in: swap one base URL (gateway), or SDK-only mode where your prompts and provider keys never leave your servers.
What I'd love feedback on:
How do you track AI cost per customer today, if at all?
Would you rather cap spend per customer, per agent, or both?
Gateway or SDK-only: which would you actually let into production?
I'll be here all day answering questions.