Workspace, Project, and Catalog Context

Hi. I wanna show you some of the new features we built around context for the Count AI agent. So I've asked an agent, a really naughty task, to tell me about ARR and it's done a really naughty agent thing. It's it's calculated based on calendar view. And so if you're work in finance or in a real business, this is probably really familiar but it's really frustrating because I want it to tell me in financial years because that's what the rest of my company operates in. This is a really great job for context. This is something that an analyst would know that a junior analyst would know but that the agent doesn't. So with context, we can take a new agent, ask the same question but give it some new information. We can tell it about how our finance system works. We can tell it about how revenue works. Show me ARR by year and we can give it the same dataset we gave the last query and we can give it these three new bits of information. And this is just like I wrote a longer prompt, but it's slightly more reusable. So we can see that it's already observing that new context and calculating by financial year. But obviously, this isn't a sustainable way of working. This is this is great for ad hoc analysis, for capturing ideas and passing it to an agent. But we want to bake some of this knowledge that's really important into how everyone works with you. So we built some new features to let you lift up this kind of canvas level context to a workspace or project level. So in our workspace settings, we now have a box for agent context and we can give it this context and this will now apply to any query whether it's a CSV targeting a database or anything else. It just has this new information. And similarly, we could apply this at project level to capture project information and to slowly override our workspace level. We've put some best practice, our line in the sand of how we think this should be used that at a workspace level, it's it's really useful to capture your tone of voice, the the terminology that's specific to your company, the the formality the formality standards, the the currency that you use, or just general information about how your company works, whether it's about KPIs, your fiscal calendar, or just how your industry works. At project level, you might want to capture demands of your stakeholders or that, you know, you care about a certain time period and you want everything in this project to be scoped to that. But context can be so much more powerful than this. The agent can also now understand your database schema when it's querying a database directly and also work with your catalog within count metrics. Let's look at where you might use this. So I have another more expansive query where I've asked the agent to examine some Google Analytics data. And there's a whole ton of this different event names that come from GA four that the agent doesn't really know anything about. So it doesn't understand how to construct funnels. It doesn't understand how to work things out. This is a great example of a problem to If I go to my catalog view, the agent now ingests the raw YAML from our catalog. So I can start just putting more information in here. YAML is YAML, so it doesn't understand multiline comments. This obviously is a great way of encoding information in a way that an analyst can look up. This is a great way of capturing information in a way that an analyst can look up and also benefits from being held under version control. I can go one step further. Account metrics now supports agents dot m d. I can write more expansive markdown formatted information and have it passed to the agent with every query. This is scoped to the catalog, so it will find the most appropriate context document to the query I'm asking. We think context is the key to getting more accurate, more usable output from the agent. And these new features let you build context that has a real impact across your work. Give it a go.