Langfuse v4.62.0

Langfuse released v4.62.0 with multiple performance, security, and feature updates across the platform: - AI Gateway : Raised request limit to 16 MiB, added configurable execution limits, implemented SecretString for credentials, added typed tenant IDs, and switched to thiserror . - OpenTelemetry & Ingestion : Added support for decoding per-key metadata on third-party spans from Langfuse SDK v5/v6 exports; derives completionStartTime from AI SDK 7 time-to-first-chunk; preserves OpenAI usage normalization when providers include non-standard fields; and handles Azure Blob 201 Created as completed uploads. - Search & Database : Fixed escaping for LIKE wildcards and literal characters ( , , % , \ ) in tracing substring search. Improved ClickHouse-like CTE name resolution across tenancy, soft-delete, dedup, and bind typing. - UI/UX & Features : Resizable panels for Topics, improved styling for the Cmd+K menu, tighter detail header rows, and updated 'Beta' label to 'Optimized'. Preserves cache epoch across prompt reads. - Audit Logs : Began staging change events next to the Postgres audit log.

• Upgrade to Langfuse v4.62.0 if using the AI Gateway to take advantage of 16 MiB request payloads and configurable execution limits.
• Verify telemetry ingestion if relying on third-party span metadata or AI SDK 7 streaming chunk metrics.