Every product has one page that is the product. Ours is Events: one row for every action any of your agents has ever taken, in order, each row carrying its own proof. Everything else in the dashboard — agents, sessions, anomalies, evidence — is a different way of reading this tape. This post is a tour of the tape itself.
One row is one action
A row on the Events page reads like a sentence: at this time, this agent called this tool, with this outcome, in this many milliseconds. Behind the sentence sits the full record — the parameters the tool was called with, the response it returned, the model and prompt context, and the cryptographic detail that makes it evidence rather than a claim. The row is the summary; the record is one click away.
What makes the page unusual isn't the columns. It's that the tape is complete and ordered. There is no sampling and no "log level" deciding what was worth keeping. If an agent did it, there's a row — the page shows your plan's live query window, and the full record stays in the durable archive for the life of your subscription. That completeness changes how you read the page: absence means something. If there's no row, the agent didn't do it — a sentence no ordinary logging stack can say with a straight face.
Finding the one row that matters
A complete tape is only useful if you can get to the right part of it fast. The page filters by agent, tool, status, and time window, and search runs across the stream — so "what did the billing agent do on Tuesday between 3 and 4am" is a ten-second filter exercise, not a forensic project. Filters compose: status Failure, plus one agent, plus last week, narrows forty thousand rows to the eleven you actually need to read. And because the counts across the top follow your filters, the page doubles as a quick analytical tool: filter to one agent and the success/failure split updates to tell you how that agent's week actually went, before you've read a single row.
The column your logs don't have
Each row shows its verification state. A verified mark means this exact record checks out against the public key you registered — the signature is your environment's, and the chain link to the previous event is intact. This is the practical difference between the tape and a log file: every row can prove it hasn't been altered since it was written.
And when something doesn't verify, the page says so, visibly. A recorder that hid its own bad news would be worthless — the whole value of a green mark comes from the fact that a red one is possible and would be shown. A chain break here is flagged exactly where you'd see it, next to the rows it affects, not buried in a weekly report. The tape also tells you how each event was captured, because not every capture path carries the same guarantee — an event recorded by the SDK at the moment of action and one collected from a platform's logs are labelled as what they are, so a careful reader is never misled about which claim a row is making.
A day on the tape
How this plays out in practice: a support lead forwards a customer complaint — "your bot refunded me twice." Open Events, filter to the refund tool, search the order number, and the two rows are on screen inside a minute: same order, nine seconds apart, both signed, both verified. Click into the second one and the full record shows the duplicate request that caused it. The conversation shifts from "did this happen?" to "here's exactly what happened, and here's the fix" — which is the entire point of keeping a tape. Investigations on the Events page tend to be boring, and boring is the compliment.
From tape to evidence
The Events page is where questions start: an odd customer complaint, a security review, an auditor's sampling request. Some questions end there — you find the row, read the record, done. For the rest, the tape feeds everything downstream: rows can be exported for your own tooling, and the same stream powers the Evidence Packs your compliance team hands over whole. The tape is the source of truth; everything else is presentation.
The best way to get a feel for it is the live demo — real events, real verification marks. Questions about what the tape captures in your stack? Ask us.