Run your own LLM proxy to see exactly what your agents and apps spend
A builder argues the provider dashboards are not real token statistics and routes every agent, harness and app through a self-hosted LLM proxy instead. One endpoint covers many providers and models, logs which key called which model from which client and when, and hands out virtual API keys so the main key is not pasted into every tool.
Source
amrrs on Hacker News · 2026-10-18 · Project link
Written in our own words. Credit to the original author. Link checked 2026-10-19.
What they did
A builder argues the provider dashboards are not real token statistics and routes every agent, harness and app through a self-hosted LLM proxy instead. One endpoint covers many providers and models, logs which key called which model from which client and when, and hands out virtual API keys so the main key is not pasted into every tool.
Result (as claimed)
Time-series usage per key and model on infrastructure he controls; the post reached 210 points on Hacker News.
Tools they used
Cost: OpenRouter: free plan, paid from Pay per token; Claude Code: from $17/mo; OpenAI Codex: free plan, paid from $8/mo.
Set it up
Pick a contained repo or task
Start on a branch or side project where mistakes are cheap.
Install and configure
Set up OpenRouter, Claude Code, OpenAI Codex. Read what it can run on your machine and limit permissions where possible.
Run it on a real task
Give it a well-scoped issue and let it work. Review the diff like any pull request.
Add guardrails
Keep tests and CI in the loop, and never let an agent push to main or touch production secrets unattended.
Copy the full setup prompt
Paste into your agent and it will walk you through the setup, asking before anything that sends, posts or spends.
You are helping me set up this AI use case end to end in Claude Code. Goal: Run your own LLM proxy to see exactly what your agents and apps spend What the original author did: A builder argues the provider dashboards are not real token statistics and routes every agent, harness and app through a self-hosted LLM proxy instead. One endpoint covers many providers and models, logs which key called which model from which client and when, and hands out virtual API keys so the main key is not pasted into every tool. Reported result (as claimed by the author): Time-series usage per key and model on infrastructure he controls; the post reached 210 points on Hacker News. Source: https://news.ycombinator.com/item?id=50145507 Tools they used: OpenRouter, Claude Code, OpenAI Codex Category: Dev & ops Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
You are helping me set up this AI use case end to end in Codex. Goal: Run your own LLM proxy to see exactly what your agents and apps spend What the original author did: A builder argues the provider dashboards are not real token statistics and routes every agent, harness and app through a self-hosted LLM proxy instead. One endpoint covers many providers and models, logs which key called which model from which client and when, and hands out virtual API keys so the main key is not pasted into every tool. Reported result (as claimed by the author): Time-series usage per key and model on infrastructure he controls; the post reached 210 points on Hacker News. Source: https://news.ycombinator.com/item?id=50145507 Tools they used: OpenRouter, Claude Code, OpenAI Codex Category: Dev & ops Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
You are helping me set up this AI use case end to end in Grok Bot. Goal: Run your own LLM proxy to see exactly what your agents and apps spend What the original author did: A builder argues the provider dashboards are not real token statistics and routes every agent, harness and app through a self-hosted LLM proxy instead. One endpoint covers many providers and models, logs which key called which model from which client and when, and hands out virtual API keys so the main key is not pasted into every tool. Reported result (as claimed by the author): Time-series usage per key and model on infrastructure he controls; the post reached 210 points on Hacker News. Source: https://news.ycombinator.com/item?id=50145507 Tools they used: OpenRouter, Claude Code, OpenAI Codex Category: Dev & ops Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
The Daily 5
Get five new use cases tomorrow morning
Free. Real sources. Unsubscribe any time.
Guides for this use case
How to vibe code an app with Claude Code · How to run multiple coding agents together · How to cut AI coding agent costs and token use
Related use cases
See all your running coding agents as a live team call
Agent Standup is a local visualizer that shows Claude Code, Codex and Mistral Vibe sessions as a live team call and session board, so you can spot the agent that is waiting on you.
Run your coding agents as pixel workers in a browser office
The Agent Office shows Command Code, Claude Code, Codex and five more agents as pixel workers. You hire them into Daytona, Vercel or Railway sandboxes, drive them from the browser and approve actions with one key.
See which private files your coding agents touched and why
Claude Code and Codex save every conversation on disk, each subagent in its own file. agentwhy reads them locally and shows which private files an agent reached, the task behind it, and which key to change, without ever displaying the secret.
Clean up the disk space your coding agents leave behind
Every coding agent task can leave a full copy of a repo on disk. The builder checked his machine and found 40 branches safe to delete and 15.1 GB of old node_modules and build folders across 62 repos, so he built Lopper to find and clear them.