In this tutorial, Mitch takes us through our new Projects feature.
What's inside a Project
A Project bundles four things. System instructions define the persona and rules. Prompts and workflows give you reusable, multi-step operations you can run on demand. MCP connectors plug in external tools like Asana, Linear, or your own services. Knowledge bases hold reference material the model reads from and output folders where the model writes results.
How conversations work
Every chat starts with the Project's context loaded. Use @ mentions to point the model at a specific prompt, workflow, knowledge base, or connector. Toggle "Run in background" to queue a task asynchronously, then return to the completed conversation whenever you want.
How Projects compound
Assign a folder as reference (the model reads from it) or as output (the model writes results to it). On the next run, the model reuses prior outputs and reference material instead of starting from zero. Share the Project with your team and everyone works with AI through the same tools and instructions.
What you can use Projects for: customer research, code review, sales prospecting, content production, ops runbooks, and anything else where the context is stable but the questions change.
Video transcript
Hey, it's Mitch here from IEL. I wanted to give you a quick demo of our projects feature which you should have access to today. Um, so projects is like a workspace for uh, you know, tasks that you repeat a lot or domain uh, areas that you work in a lot where you can connect chat and prompts and workflows and knowledge bases as tools um, and you know uh, really start to find some efficiencies. It's also where we are starting to introduce the concept of agents and sub aents for uh, for your work. So to get started, you can see here on the lefth hand bar, you have a new projects tab, uh let's create a new project.
For this one, I'm going to create a market research project. Um this is just the configuration page. You would set the default model that you would be chatting to. I'm going to use the new Claude OBS47. You can set some advanced uh settings in here as well.
Cool. So every chat uh is comprised, sorry, every project is comprised of chat tools, which we'll configure in a second, and knowledge bases. Let's configure the knowledge base. So I have here a research knowledge base that I'm going to set as reference and output. Reference means that whenever I ask chat a question, it's first going to check what files I have in the knowledge base.
Uh and it will look for an answer in there. Uh output means if I create an artifact, uh that will be the place that it will store it. I'm now going to set up some tools. Right? So tools uh are comprised of all of the things that you would use for today.
So we can connect uh prompts and workflows that you might already have, scripts that you might have already built, uh external MCP that you might already connect to. So different tools that you might use uh that you want to loop Aisle into. Um for this one, we're going to connect a company identifier prompt and a market research prompt. System tools are always configured, but you can turn them off. This is just web search and web fetch, meaning that this project can uh get content from the internet.
Cool. Uh and so to start, I'm going to say what reports do we already have in our research folder? Right. You can see that it's pulled the three reports that we have in our folder. I want to add two new reports for two different industries.
So, I'm going to say uh create and save two new uh reports for the following industries in 2026. Let's say financial services and the automotive industry. And you can see here that what the project chat is doing is it's spinning up two separate agents that are going to work outside of the context of this chat um and achieve these tasks um in parallel. It is also going to use the prompts that I'd already created that do the market research in the way that I like. So both of those things are going to happen in the background and then be returned to this chat once they're saved in uh the knowledge base that we' specified.
Okay. And so both of these reports have been generated and saved to the knowledge base. So we have five reports in the knowledge base. Um why don't we ask a question uh of the chat across all reports. So all industries covered in the knowledge base.
What are the key challenges that they share? What are the biggest risks in 2026? And so now this is doing you know uh just traditional rag. It is listing what reports are in there. It's looking at the propositions that would have been generated by that knowledge base and it's doing quick cross industry analysis.
Right? So all five industries geopolitics and tariff stuff specific stuff about AI. You know pretty much all the things that you would expect to see. We can now call out a specific tool that we might have listed in there. So if we say uh company identifier what are the companies that are most vulnerable to let's say geopolitical supply shock in the automotive industry.
Right now it is using that prompt which again uh we connected as a memory here launching the tasks as a sub agent and uh that prompt is returned these companies and rank them by who is the most uh at risk for for that area. Some of the other cool things about projects you might have spied it before but you can launch things in uh the background. So let's say we wanted to just launch um the market research assistant uh create and save a new report on the I don't know uh hospitality industry in 2026 you can launch that in the background and so that will just happen in the background once that port is ready it'll be available in the knowledge base that we configured uh before uh so this is just a quick look at um our projects feature. Uh we're just scratching the surface. I really hope you get your hands on it.
Um and you know, let us know of any feedback as you use it.