See what a conversation costs, and speed up chat with an AI Classifier

Chat now shows what it costs. A subtle running total sits under the model selector and in the context-usage popover, so you can see the $ cost of a conversation at a glance instead of guessing from token counts. It's priced automatically for OpenAI, Anthropic, and Google models using each provider's own published rates — no setup required.

You can also add an optional AI Classifier key to speed up chat. Before calling your main model, SupaPM asks a fast, cheap routing model which category of tools a message likely needs, and only sends that category instead of every tool every time. It's a bring-your-own-key connector with the same local-only privacy model as your other provider keys, and it's entirely optional: without a key, chat behaves exactly as it always has.

What's new

• Sidepanel chat shows a running "Conversation cost" estimate under the model selector and in the context-usage popover, covering the whole chat, not just the latest message.
• Cost is estimated automatically for every current OpenAI, Anthropic, and Google model, using each provider's published per-token rates, including the discount for cached input and the extra charge for cache writes.
• Settings → AI Configuration lets you see and override the price for any model, right where you manage that provider's API key.
• Settings → AI Configuration has a "Test key" button for every provider (including Custom endpoints) that makes a live check to confirm a key actually works before you rely on it.
• New optional AI Classifier connector: add a Jev API key in Settings to let SupaPM route each message to only the tools it likely needs, reducing cost and latency on turns that don't need everything. Skip it and nothing changes.