I’m tired of starting my day by opening a dashboard and trying to work out what happened yesterday.
Anyone who owns a metric knows the routine: change the date range, compare with last week, look for something that deserves a closer look. If a number seems wrong, open another tool. Then the logs. Then the customer records. Eventually, there are enough pieces to start an investigation.
The frustrating part is not reading the chart. It is that the whole process waits for someone to remember to look.
That frustration is part of why we started Altertable. We wanted our data to be investigated continuously, not just stored until someone had a question.
Today, we’re introducing **[Cousteau](/product/cousteau)**, our always-on investigation agent. It follows changes across your whole business, from revenue and retention to feature adoption and production errors, tests explanations, and brings back findings with evidence. It works in the background, without waiting for you to start a conversation.
## Your attention should not be the scheduler
Putting a chat box in front of a warehouse makes it easier to ask questions. But you still have to think of the question.
You have to remember the feature you shipped two months ago. You have to notice that a particular group of customers is behaving differently. You have to make time to check whether the improvement you celebrated at launch actually lasted.
Meanwhile, your team has moved on. The release is still in production. Customers are still using it. Its effects do not stop because the project is no longer on your roadmap.
_Cousteau keeps looking._
You can give it particular areas to focus on: onboarding, checkout, adoption of a new feature, or the behavior of an important customer segment. Whether you run the business or build the product, you set the priorities without having to anticipate every question or write every query.
That attention is useful for more than catching regressions. A new route through the product might be working unexpectedly well. A feature might be finding an audience outside the group it was designed for. A slow change might deserve attention before it becomes an obvious problem. And sometimes the cause is outside your product entirely: a competitor’s launch, a holiday in one market, or a news event can change how customers behave without anything in your release log explaining it. When the data points outside, Cousteau can search the web to check whether something happened.
## Get signals without the noise
Always-on should not mean always notifying. An alert that says “conversion is down” still leaves most of the work to you.
Cousteau runs on Altertable’s **[context lakehouse](/blog/2026-08-14-the-context-lakehouse)**, where it can join product events, logs, traces, and business data. Some sources are ingested in real time; others are replicated or queried where they already live through federation. The explanation can span your stack without you piecing it together between tools.
Take a hypothetical checkout drop. Cousteau can compare browsers and subscription plans, join failed sessions to traces, and inspect the relevant release and code changes. It runs more queries to test possible explanations before bringing back a finding:
> Checkout completion dropped 11%, concentrated among Safari users on annual plans. The affected cohort includes 83 accounts.
>
> Failed sessions show a sharp increase in `PaymentIntentTimeout` traces. The failing code path was changed in release 2.41.
Instead of a chart that moved, you get a specific cohort, a measured impact, and the traces and release to look at next. It is still a lead for your team to confirm, but one that arrives with its evidence attached.
Your [metric definitions, product knowledge, and previous findings](/blog/2026-07-22-hand-the-agent-a-map) give Cousteau context to interpret what it sees. More relevant data and context give it more ways to test an explanation.
When nothing meaningful has changed, it stays quiet. Most runs end that way. Cousteau still checks things: a dip in signups that matches a public holiday, a spike in errors from a single test account, a jump in sessions that the same week last year also had. It records what it checked and why it dropped each lead, so you can review its reasoning without being interrupted by it.
## Built for thousands of questions a day
An investigation can involve dozens of queries, with each result guiding the next. Repeat that work throughout the day and you are building for thousands of queries, not a handful of dashboard refreshes. They need to be fast and affordable enough to keep testing hypotheses.
That is why so much of our recent work has gone into the engine itself. Most of the questions Cousteau asks are behavioral: where users drop out of a funnel ([MATCH_RECOGNIZE](/blog/2026-08-07-match-recognize-that-scales)), how their sessions changed ([SESSIONIZE](/blog/2026-09-01-session-windows-in-duckdb)), which paths they take instead of the one you designed ([JOURNEYS](/blog/2026-09-17-journeys-in-duckdb)). We added operators for exactly those questions, and run them inside DuckDB’s execution engine, so an agent can ask them hundreds of times a day without waiting on a warehouse. The same engine ingests product analytics events and OpenTelemetry logs and traces in real time, and federates the warehouse, databases, and business tools you already have. So a single query can follow what users did minutes ago, the traces behind it, and the accounts it affected, with no pipeline to build first and no nightly sync to wait for.
Some questions also reach far back. To tell whether a feature shipped six months ago is still paying off, Cousteau needs six months of detailed events, not a 30-day sample. Altertable keeps that full history queryable and runs each investigation on its own compute, so Cousteau can dig as deep as it needs without slowing down anyone else’s dashboards.
## What are you missing while nobody is looking?
A release nobody checks anymore. A customer segment quietly struggling. A new behavior that could change what you build next.
These are the things we hope someone will notice between meetings. They deserve more than whatever attention is left over.
Getting started takes one source. Connect the warehouse you already have, or start sending product events, then tell Cousteau what you care about most: onboarding, checkout, the launch you never got around to checking.
The next time you open your laptop, the investigation should already be underway.
**[Put Cousteau to work](/signup)**