None of this is implemented. This page describes the plan, not a shipped capability, so it can be designed around honestly, and so nobody mistakes it for something live.
Your agents, analysed by ours.
Reading every conversation does not scale, and the ones worth reading are the ones you would never think to open. We are building a layer that reviews them for you, surfacing what a human reviewer would catch, if a human reviewer could keep up.
Capabilities are modular by design, priced by what you pick, more on that below. Every row here is a proposal, not a commitment to ship it as described.
Eight ways of reading data you already have.
Automated insight digests
A standing summary of what changed, instead of a dashboard you have to remember to check.
Anomaly detection
On error rate, latency, cost and escalations, flagged before a client notices.
Conversation quality scoring
Every exchange rated, so the ones worth reading rise to the top.
Failure & root-cause clustering
Similar breakages grouped together, instead of one ticket at a time.
Prompt & tool recommendations
Concrete suggestions, tied to the failures that motivated them.
Topic & intent mining
What your users are actually asking for, aggregated across every conversation.
Cost-optimisation advice
Where spend is going and what would bring it down, model by model.
Natural-language querying
Ask your own data a question instead of building a chart for it.
This is why the modules above are plausible rather than aspirational, the raw material already exists.
It reads the same store the dashboard does.
Every conversation is already kept as structured data, messages, tool calls, model calls, timings, tokens and cost. Nothing about this layer requires new instrumentation from you; it is a second reader of the data your agent already produces.
Quoted, not priced, because the module list above is exactly that: a list you pick from, not a package.
Quoted by module, because capabilities are modular.
You choose which modules you want and at what volume, and it is priced against that, not a flat add-on to a plan you may only use part of. It sits on top of the Custom plan.
See where it sits in pricingAI agent analytics questions.
Is any of this live today?
No. Every module on this page is a proposal, not a shipped capability. Nothing on the rest of the site depends on it.
Why not just build one flagship feature and ship it?
Because customers do not all want the same slice, some want anomaly alerts, others want a natural-language query box. Shipping it as modules means you only pay for what you use.
How do I get early access?
Join the waitlist from this page or book a demo. We are prioritising customers already using the analytics dashboard, since the value here comes from data you already have in AgentSight.
Will this change what data you collect?
No new instrumentation is planned. It reads the conversation, message, tool and usage data the SDK already captures, model spans still carry accounting only, never prompt or completion text.
Get on the list before it ships.
- Full dashboard on the free plan
- No card required
- Model spans never carry prompts