PostHog

CUPED variance reduction landed for Experiments, giving more statistically sound results without manually controlling variables.¹ The Logs view gained saved views, and data warehouse tables now work inside funnel and lifecycle insights.²

LLM Analytics got cluster and sentiment analysis — LLM sessions can now be grouped into thematic clusters and scored for user sentiment, making it easier to spot patterns across conversations at scale.³

Error Tracking expanded its language reach — iOS SDK and React Native JavaScript support joined the existing backend integrations; weekly digest emails and Sentry as a data source also arrived.⁴

Feature flag events now deduplicate per group — posthog-node fires $feature_flag_called independently for each group combination, so flag analytics stay accurate in multi-tenancy scenarios.⁵

Python SDK tracks OpenAI structured outputs — chat.completions.parse calls are now captured, extending LLM observability to the structured response path alongside existing streaming and completion tracking.⁶

posthog-cli cleaned up its supply chain — v0.7.13 drops axios, follow-redirects, and related transitive deps that carried open CVEs; v0.7.12 added --skip-on-conflict and --force flags to symbol upload commands.⁷⁸