Launched this week
ProxAI
Hard budget caps for LLM APIs — one base_url change
5 followers
Hard budget caps for LLM APIs — one base_url change
5 followers
ProxAI is a transparent reverse proxy that enforces hard hourly/daily spending caps per LLM API client. Change one base_url — no SDK required. Works with OpenAI, Anthropic, Mistral, Groq, DeepSeek, and 5 more. Over cap → 429 before the upstream is ever called. Worst case is your cap, not your credit card. Self-hostable: single Go binary, ~10 MB Docker image (amd64/arm64), no telemetry. Or use the hosted beta at proxai.eu — free, no credit card. AGPL-3.0. Built end-to-end with Claude Code.
API cost becomes painful surprisingly fast. Curious—what pattern do you see most often when teams realize they actually need ProxAI instead of just monitoring usage manually?
@aryan787544 Ha, "surprisingly fast" is the whole thing
Monitoring is a rear-view mirror — you see the spend after it's gone. The moment people reach for ProxAI is always the same: a retry loop or runaway agent quietly torches the budget over a weekend, and the dashboard only tells them Monday.
So instead of drawing a scary graph, ProxAI just returns a hard 429 at your cap. Stops the bleeding in real time.
Fresh launch so I'm speaking to the problem I built it for more than a big sample — but does that weekend-surprise match what you've hit?
@matthieu_chartier1 That distinction between monitoring and prevention is interesting.
Feels like there's more to unpack than fits in a Product Hunt thread. If you're open to continuing the conversation, what's the best email to reach you?
@aryan787544 Glad the monitoring-vs-prevention angle landed — agreed it's deeper than a PH thread. You can reach me at j574nke2c@mozmail.com. Drop me a note on what you'd like to unpack and I'll come back properly.
@matthieu_chartier1 Thanks! I’ve just sent it over.
Looking forward to hearing your thoughts whenever you have a chance.
Hi Product Hunt! I'm the maker.
I built ProxAI after reading too many postmortems of agent loops and leaked keys burning API budgets overnight — while provider alerts are monthly and org-wide. It looked like a problem that deserved a hard, technical answer.
Two things I want to be upfront about:
1. I never got burned myself — I built this to prevent it, not recover from it.
2. This project is an experiment: it's built end-to-end with Claude Code, without me writing a single line of code. I specify, review, test and decide; the AI writes. The commit history shows it plainly.
Happy to discuss that workflow as much as the proxy. The self-host path is first-class: docker compose up, multi-arch image on ghcr.io, no telemetry, no account with me.
Happy to answer anything — would especially love feedback from anyone running LLM workloads in prod.