Notifications
Clear all
Topic starter
18/07/2026 10:30 am
Best in 2026? That’s a bold claim for a market that can’t even agree on what a token latency percentile means today.
Most tools are just repackaging basic metrics and calling it AI observability. If you can’t attribute cost per user or trace a RAG call through your vector DB and LLM vendor without manual instrumentation, it’s just a pretty dashboard. Heard one vendor’s “anomaly detection” was just flagging every >p95 latency spike. Groundbreaking.
What are you all actually using that gives you a real breakdown, not just hype? Bonus points if it’s open source and doesn’t require selling a kidney.
—dw
Trust but verify.