AI Tracing Structured ai_trace records are now emitted for AI calls across Vercel AI SDK, OpenAI, Anthropic, and OpenRouter. Traces include provider, model, usage, finish reason, timing, and first-chunk latency for streams — all correlated with the surrounding HTTP request automatically. Content capture (prompts, outputs, tool payloads) remains opt-in.
End-to-End Trace IDs A Blyp trace ID is now generated per request, returned via the x-blyp-trace-id response header, and propagated into downstream logs and database records. Covers all framework adapters: Express, Fastify, Hono, Elysia, NestJS, Next.js, Astro, Nitro, React Router, SvelteKit, and TanStack Start.
Durable Connector Delivery Queue Connector forwarding now supports an in-memory + SQLite-backed queue with configurable retries, backoff, and overflow strategy — no Redis required. Opt in with connectors.delivery.enabled. Queue state is written to .blyp/ (auto-added to .gitignore).
Dead-Letter & Queue Inspection Exhausted delivery jobs are moved to dead letters instead of silently dropped. New helpers expose queue status, dead-letter listing, retry, and clear operations for Studio and custom tooling.
---
---
Automatic Redaction Sensitive values are now sanitized before reaching any Blyp sink — console, file, database, connectors, and client ingestion. Covers common key names (passwords, tokens, cookies, API keys), bearer tokens, JWTs, and Luhn-valid card numbers. Extend with redact.keys, redact.paths, or redact.patterns.
---
---