`Voker`: agent analytics SDK around `intent`, `correction`, and `resolution`

Most teams still learn agent failures from user complaints. Voker turns conversations into intent/correction/resolution metrics; the framing is worth stealing even if the product skews team-heavy.

• —The pain is real: 90%+ of surveyed YC founders said they only notice production failures after customers complain.
• —Instead of raw traces, Voker models each conversation as intent, correction, and resolution, which maps closer to product outcomes.
• —The SDK is LLM-stack agnostic, so it positions itself as an instrumentation layer rather than a full platform rewrite.
• —Uploading logs into Claude or ChatGPT for summaries is the common workaround; this launch suggests that gap is now big enough for a dedicated product.