Preduce adds inline prompt-compression playground

Prompt compression is starting to look like real infrastructure, not just a benchmark trick. The interesting bit here is the recompression-every-turn design: in long-running chats, savings compound instead of tapering off.

• –Long system prompts are the highest-leverage target, so this is most useful for agent workflows, support bots, and other prompt-heavy apps.
• –Re-running compression on the full context each turn is the right pattern if you want costs to stay bounded as conversations grow.
• –Supporting both OpenAI and Anthropic models makes this feel like provider-agnostic middleware rather than a one-off demo.
• –The 10-message public playground is a smart growth move, but it also suggests this is still early-stage productizing rather than hardened production infra.