Jared Siegel wants to know which client called which model, with which key, and what happened next. He argues for running an LLM proxy, a service that routes requests between coding tools and model providers. His case goes beyond the usage dashboard:

  • One endpoint for several providers and models, making switches easier.
  • Request logs showing the client, key, model, time, and status.
  • Separately issued virtual keys, keeping the main provider secret out of individual agents.

Siegel says his setup lets him diagnose a coding session’s cache miss in 20 seconds by hand. He leaves the proxy unnamed and reports his own experience, without a comparative test. The engineering argument is straightforward: own enough of the request path to investigate failures.

The 152-comment thread on Hacker News adds alternatives and costs the essay leaves underdeveloped.

What the thread adds

  • CharlieDigital — “You don’t need a proxy to do this.” They propose OpenTelemetry, a standard for exporting diagnostic traces: “It is less intrusive and does not become a single-point-of failure since the OTEL emission and collection is best-effort whereas a failing/overloaded proxy is going to strand your team until resolved.”
  • jakozaur — “For subscriptions, it is easier to collect the JSONL files from ~/.claude/projects/ and similar directories. They are better organized (e.g., subagents) than what the proxy sees.” On collecting sessions across an organization: “You’ll need to scrub credentials and handle security.”
  • lionkor — on personal information in conversation logs: “If you’re in the EU, be aware that the chats might contain PII.” Their advice: “So id be very careful with chats and treat them like chats, not like code.”
  • NeutralWanted — on metered API costs versus a subscription: “I tested it once, I’m in on the 20x claude max ($200/month) and I either max my weekly limit every week or come really close. But it only took 2-3 days to hit $200 spend with api keys.”

Handles are pseudonymous; HN publishes no per-comment scores, and its ranking supplies the sampled order. This is a slice of the thread, not a consensus.