Transparent usage-billed LLM API gateway after subscription and markup fatigue
A heavy OpenCode/Claude Code user describes burning through a $200 MAX subscription in about an hour, then getting burned by relay stations with opaque billing (one charged $7 for a single Opus task; another drained a top-up in a few embedding calls with bills that didn't add up). They settled on a gateway with per-token pricing and an auditable usage ledger — evidence that billing transparency, not price alone, is the switching factor for AI API intermediaries.
The problem
Developers using coding agents face unpredictable subscription burn and relay stations whose billing is opaque or inflated. The pain is not access to models but trust: usage details must be queryable in real time and reconcile with actual token counts, and one key should cover many models.
What could be built
An OpenAI-compatible multi-model gateway competing on verifiable billing: real-time usage ledger, per-request cost breakdown, token-count verification endpoint, one key across GPT/Claude/Gemini/DeepSeek. Target agent users whose subscriptions run out mid-month.
Who it's for
Who is talking（1）
Related topics
Summaries are AI-generated. The original words are in the threads above.