OpenObserve
AI-native, open-source Datadog alternative
Start new thread

We're launching AI Observability this Thursday. Here's what it is and how you can help.

by
•

Hey everyone,

Quick heads up: OpenObserve AI Observability goes live on Product Hunt this Thursday, September 10, at 12:01 AM PT.

What we're launching:

If you've built anything with LLMs or agents, you know the debugging story is rough. Your APM tool sees the HTTP calls. Your LLM tracing tool sees the prompts. Nothing sees both, and nobody can tell you why the agent gave a bad answer or which hop in a five-agent workflow burned the latency.

We're fixing that by putting AI observability inside the same platform that already handles your logs, metrics, and traces:

  • Monitor: Sessions, Agent Graph, Agent Behavior, and Insights. Every run, every tool call, every hop, with latency and cost attached.

  • Evaluate: Scorers and Eval Jobs that run against production traffic, so you catch quality drift, not just error spikes.

  • Annotate: Review queues and datasets built from real traces, so humans can grade what the scorers flag.

Same Rust engine. Same OpenTelemetry pipeline. Self-host it or use OpenObserve Cloud. And because it all lives in one place, you can trace an agent's bad output straight back to the slow database call underneath it.

How you can help on Thursday

We're aiming for a top-3 finish, and that comes down to real engagement in the first few hours. Three things make a difference:

  1. Hit "Notify me" on the teaser page now so you get pinged when we go live:

  2. Leave a comment on launch day. A question, a use case, what you'd want to see next. Honest feedback beats "congrats," and we'll reply to every single one.

  3. Share it with anyone on your team who's shipping LLM features and still debugging by reading raw prompts in logs.

We'll post the live link here the moment the page is up.

Thanks for being part of this. A launch is only as good as the people who show up for it, and this community has shown up every time.

See you Thursday.

Jake and the OpenObserve team

150 views

Add a comment

Replies

Best

The eval loop with human review queues integrated is cool! Many teams find it tough to connect production logs with real quality scoring. I'm really looking forward to trying this out on Thursday!

I really appreciate that you’re sticking with the same Rust engine and OpenTelemetry pipeline instead of just throwing together a separate sidecar. It’s a huge relief not to have to juggle another isolated platform just for agent metrics. it really makes a difference!

the join is the right problem. the part i would want to know about is the unit.

apm thinks in spans, and a span is a call. agent debugging needs decisions, and a decision does not map onto a call. one decision can be four tool calls or zero, and the interesting failure is usually the zero. the agent decided not to look something up, and there is no span to hang that on.

so when you say every run, every tool call, every hop, my question is whether a hop that did not happen is representable at all. most bad answers i have chased came from a lookup that was skipped rather than one that was slow, and a system built on emitted spans has no row for the call that was never made.

the annotate half is the bit i think is underrated in the post. scorers drift with the thing they are scoring. a human queue built from real traces is the only part of that stack with an outside view.

good luck thursday. we are on the same board.

Can't wait for the launch. Agentic observability is another layer in a bigger observability stack and should not require another tool or separate instrumentation. That's what led us to the tool sprawl that we are seeing these days.

Really excited to see this go live. As AI agents become part of everyday production systems, their observability should fit naturally into the tools teams already use, not force them to adopt another disconnected platform. Adding separate tooling for every new layer is exactly how observability stacks become fragmented and difficult to manage.

And we are now live! We would very much appreciate your questions, comments, and general support