I built something....

So, I built something. This is Journey.

It's something that I've been working on for a few weeks and I wanted to tell you about it. I'm looking for feedback. So,

about it. I'm looking for feedback. So,

drop it in the comments below. So, the

problem that I am solving with Journey is the fact that once you've built out a workflow for an agent, it's really hard to give that to other agents. And if you

have your own agent, it's really hard to discover workflows. When I was making

discover workflows. When I was making all those videos about Open Cloth, the question I got the most is, "What are you actually using it for?" And so, the use cases, the things that I was doing

with it were the most interesting to everybody who was watching those videos.

And so, I realized it's actually kind of difficult to figure out what other people are doing with their agents. And

if you wanted to do the same thing, it's even more difficult to replicate it. And

so, that's where Journey comes in. You

can think of it like a registry for end-to-end workflows and we call those kits. A kit is a fully packaged workflow

kits. A kit is a fully packaged workflow ready to install for your agent. So it

includes skills, it includes tools which are regular code. It includes learnings, memories, the services that you've used with it, tests, failure examples, and so

much more. all packaged into this single

much more. all packaged into this single installable kit so that your agent doesn't have to reinvent the wheel from scratch every time. So, Journey is the

product. It's the registry and kits are

product. It's the registry and kits are those packages. You can kind of think of

those packages. You can kind of think of it like npm and I tried to make it as easy to install these workflows as possible. You should literally just be

possible. You should literally just be able to point your agent at it and it will download everything, know how to use it immediately, know what edge cases there were, and anything else. I also

spent a good bit of time thinking about how teams of people and teams of agents might use this. So, let me show you all that. So, this is my knowledgebased rag

that. So, this is my knowledgebased rag system kit. And this is a workflow that

system kit. And this is a workflow that I use all day, every day. Anytime I come across an article, a tweet, a video, a paper, anything at all, I throw it into

Telegram. It gets ingested into this big

Telegram. It gets ingested into this big database and then my agents can reference it later. So if I'm making a video in the future and I think, oh, I should probably reference all the things

that the Claude code team have been building in the last few weeks, I can ask in natural language to my agent, hey, show me all the features that the Claude team has released over the last few weeks. It will look through the huge

few weeks. It will look through the huge database of articles. It will find all the tweets cuz I've saved them all. And

then it will show me. And I can use this in a bunch of different ways. When I'm

making a video about it, I can have it automatically incorporate anything that I've looked at in the past into the video outline. And there's a lot of

video outline. And there's a lot of reasons that I can personally use it, but I realized having this knowledge base is incredibly valuable for not only me, but for the rest of the people on my team. I want them to be able to throw

team. I want them to be able to throw articles that they come across into this database. And I want them to be able to

database. And I want them to be able to use this workflow. And also maybe you maybe you would gain value from it and I want to share it and I want a way for you to install it easily without having to recreate it all from scratch. This is

the install and I'm going to come back to that. But the kit itself includes a

to that. But the kit itself includes a few things. When a kit gets published,

few things. When a kit gets published, we have first dependencies. That's

everything that the kit requires to run properly. So, first we have an anthropic

properly. So, first we have an anthropic API key. That's just the model I used,

API key. That's just the model I used, but you don't have to. You can use any model you want. Then we have node. We

have the summarize CLI. And I'm using OpenClaw. Now, of course, if you're not

OpenClaw. Now, of course, if you're not using OpenClaw, if you're using Nemoclaw or if you're using Claude Co-work, it doesn't really matter. When you install a kit, it should adapt itself to

whatever agent environment you're using.

Here we talk about the models. So it is verified with claude. And we also have an embeddings model. Now you can use any embeddings model you want. You can use a local embeddings model. You can use

OpenAI or Google or Olama Nomi which I used Nomic for a long time. Then it also includes external services. So to parse tweets properly I'm using FX Twitter. As

the web scraper, I'm using Firecrawl.

And as the browser, I'm using Chrome DevTools. So, as you can see, it's

DevTools. So, as you can see, it's starting to put together this picture where once you show it to your agent, your agent will know exactly how to use this. It also includes failures

this. It also includes failures overcome. So, these are the different

overcome. So, these are the different problems that I have solved over time with my agent to make this work really well. I could just give you a prompt,

well. I could just give you a prompt, and that's how I've done it in the past.

I say, "Recreate it using this prompt."

Now, you're going to be burning all of those tokens. you probably won't have an

those tokens. you probably won't have an identical set of features to what I have and you're gonna have to spend your own time figuring out how to get it to work, but now hopefully you don't. We also

have this content tab. We have the kit.md file. You can click into it and

kit.md file. You can click into it and you can start to read what it is. So, we

have a goal, when to use it. This is

just a markdown file, natural language, how to set it up, the environment that it runs in, how to get it working. So

there are the steps, some failures, we overcame, some validations, some outputs and source files and so on. We have the knowledgebased skill. So remember the

knowledgebased skill. So remember the kit is the whole thing. It's the package of everything so that you can just install it. Now we have this

install it. Now we have this knowledgebased skill which tells your agent exactly how to operate the knowledge base. The database schema. So

knowledge base. The database schema. So

this is exactly what our database used and you can use it as well. Now

obviously you can customize it any way you see fit. And we have a bunch of source files. So this is traditional

source files. So this is traditional code that your agent can use aka tools to actually run the knowledge base. And

this all installs in one go. It's very

easy. Another thing about kits is we can version it. So you can have a version

version it. So you can have a version that you've used from me and then I can update it and you don't have to update immediately. You'll get notified. Your

immediately. You'll get notified. Your

agent will actually ask you, hey, there's a new version. Do you want to update? Here's what changed. Here's

update? Here's what changed. Here's

what's better. Every single version is here. And then learnings. Now, there are

here. And then learnings. Now, there are no learnings here yet, but obviously since every single agent, every single model, every single environment is going to be different. It's not always going

to work smoothly. So, what I wanted to do is have a central repository in which agents can tell you their learnings like, hey, I'm using this version of

Node or hey, I'm actually using GPT5.4 and it works fantastic. All of those learnings can feed back in and then your agent can read from them and these kits

become better over time. So that's a kit. Now, how do we actually install it?

kit. Now, how do we actually install it?

Well, when I first started making this, I wanted to make it agent first. This is

a product built for agents. The further

I got along, I realized I did need to expose some of this to the humans responsible for the agents because you probably wouldn't install a bunch of code or a bunch of markdown without actually seeing what's in it. But the

install is still very agent first. If

you're using something like OpenClaw or any kind of modern agent, you can pretty much just copy this prompt and tell your agent to install it. So fetch the journey kit from and then here's the

URL. And of course, Journey itself is a

URL. And of course, Journey itself is a kit and follow it. So you can simply click copy prompt, give this to your agent, whether you're using OpenClaw, Nemoclaw, Hermes agent, whatever it is.

As long as it has access to an environment, you should be good to go.

You can even give this to Cloud Co-work and Cloud Code and Cloud Code Desktop.

It should work with all of these different harnesses. And if you want an

different harnesses. And if you want an even more customized implementation, you can install it via CLI right here. npm

install-g journey kits. And we have a bunch of other options here as well. So

cloud desktop, I actually built a very specific guide for how to install it in cloud desktop. All of this is available

cloud desktop. All of this is available from the homepage. All right. So, like I mentioned, I really tried to make this agent first. Once you install the

agent first. Once you install the Journey kit, your agent will know exactly how to use Journey. You actually

don't really ever need to go back to the website for most of the features. When

you start getting into the team features, it's okay. It's actually

pretty useful to go back to the website, but again, you don't really need to.

Everything can be done through your agent. So, let's try searching. So, show

agent. So, let's try searching. So, show

me a kit that'll help me code better. I

hit enter. It's going to process my prompt. figure out that it needs to use

prompt. figure out that it needs to use the journey kit, the journey skill, and then go ping the server. Find any kits that are relevant to coding and give me the top one. So, here we go. Best match

for code better is Matt Claude, which is the publisher. That's my agent. Code

the publisher. That's my agent. Code

refactoring planner v1. Great. It tells

you what it does. So, analyzes a codebase for refactoring opportunities using static complexity metrics plus cloud. Produces a prioritized phase

cloud. Produces a prioritized phase refactoring plan. So these can be

refactoring plan. So these can be simplistic workflows but they can actually get very sophisticated. One of

the most sophisticated is that knowledgebased rag system and then all you have to do is say install it and that's it and it will install that kit end to end. Obviously if you want to learn more about it you can go to the

website you can ask your agent whatever you want to do. If we go to browse kits you can see a bunch of different kits.

All right so here's another kit that I published that you can use immediately and by the way this entire thing is free. I do have some team features. I

free. I do have some team features. I

have some enterprise features. Maybe in

the future, I'll think of some kind of pricing model for that, but for the time being, it is completely free, especially for discovering and installing kits on your own. Go ahead and use it. Here is

your own. Go ahead and use it. Here is

the weekly earnings preview. One of my favorite. Every Sunday, I get a list of

favorite. Every Sunday, I get a list of all the stocks that it thinks I might be interested in that are going to have earnings calls that week, and it asks me, "Which ones do you want me to summarize for you?" Then I give it the

list and then every single day after the earnings call, it summarizes the earnings call and sends it to me. It's

really neat. So I publish this kit and we have this release review. I am doing an entire analysis of each kit which was really fun to build and so I rank it on

security. So this is a 7 out of 10

security. So this is a 7 out of 10 completeness which is how easy would it be to install, how complete is the package and then how difficult is the overall setup. You can click in, you can

overall setup. You can click in, you can see who the kit is best for, different things to watch out for. And just like before, we have all these other fields.

Now, to install this specific kit, you can copy the prompt right there, or you can click here and install it via CLI.

And publishing a kit is super easy. And

this is actually something that I'm going to ask from you all. If you're

playing with this, if you enjoy it, please publish your workflows. other

people might get value from them and it's fun to see other people use it, give you feedback. The workflows, the kits get better over time and it's just awesome to share your learnings with the community. All you need to do is

community. All you need to do is describe the workflow in just a few words and say publish it as a kit and it will know. It has all the instructions

will know. It has all the instructions and how to package it up and publish it appropriately. Then I'll do the analysis

appropriately. Then I'll do the analysis on my end for security, for completeness. I'll give you

completeness. I'll give you recommendations on how to improve it if any. And then that's it. It's actually

any. And then that's it. It's actually

very straightforward, very simple. You

need to do two things to publish. Sign

up and you need to verify your email address. That's it. No payment, no

address. That's it. No payment, no credit card. It's all free. Go ahead and

credit card. It's all free. Go ahead and publish. So, in the top right, there's a

publish. So, in the top right, there's a sign-in button right there. Registration

is super simple. All you need is an agent name. That's it. You can sign up

agent name. That's it. You can sign up with an email, but it is optional.

However, to publish and to use the team features, you do need to verify an email address. So once you register, you're

address. So once you register, you're going to get a key for your agent. Then

you can use that to sign in. And this is what your console looks like. Here you

have your dashboard with the number of published kits, the organizations you're a part of, which is basically like a team. You have your keys, your

team. You have your keys, your permissions. Here are all of your recent

permissions. Here are all of your recent kits. Here are the organization specific

kits. Here are the organization specific kits. And I'm going to explain how that

kits. And I'm going to explain how that works. It's really cool. And you can see

works. It's really cool. And you can see all the settings on the left over here.

So let me show you organizations. Now,

one thing I wanted to be able to do was take a kit or an entire workflow that I had with my agent and allow my team to use it. I didn't want them to just

use it. I didn't want them to just invoke my agent because agent systems like OpenClaw were not meant to be multi-user. Plus, I didn't want to

multi-user. Plus, I didn't want to accidentally leak private information.

So, of course, they should have their own agent. And so, I think I solved a

own agent. And so, I think I solved a way for us all to use the same workflows and shared context while they all are using individual agents. So let me show you this is forward future. This is my

company and we have other agents on the team. You can add agents easily. You can

team. You can add agents easily. You can

create different teams easily and you can set the permissions for each agent on your team. Very very easy. You can

also invite other people to your organization. But there's one major

organization. But there's one major problem. I've already built up my

problem. I've already built up my knowledge base over the last few months.

It has hundreds of articles in there already that I want them to have access to. I don't want us all to have separate

to. I don't want us all to have separate databases and how would our agents talk to each other? It just would be very messy and so that is where shared

contexts come in. It allows our agents to all point to a shared set of context or credentials. So for example, we use

or credentials. So for example, we use one password at my company and we're using a hosted database to store the knowledge base. All the other people on

knowledge base. All the other people on my team install this kit and as part of the kit, it tells it where to find the credentials for the database. But here's

the thing. Journey does not store the credentials. It just points to the

credentials. It just points to the system that does. So again, we use one password. I say exactly where to find

password. I say exactly where to find the credentials and one password and then whoever else installs this kit, their agent will know exactly how to set it up and how to find the existing

database. So you can find it here. If we

database. So you can find it here. If we

click into kits, this is my specific kit. It is not public. It is a private

kit. It is not public. It is a private kit. It is a kit specifically for my

kit. It is a kit specifically for my organization. I can fork off of public

organization. I can fork off of public kits and make them private for my organization. I can create my own

organization. I can create my own private organization kits. All of this is very easy. And so we scroll down and we can see here org resource bindings.

So we have a superbasebacked knowledge base with 368 sources. and this many chunks. The credentials are stored here

chunks. The credentials are stored here and it basically explains exactly where to find the credentials. We also use the firecrawl API but I have not created the shared resource. And so since firecrawl

shared resource. And so since firecrawl was identified as something that could be shared context, I can easily just create it. I click create resource and

create it. I click create resource and then it prefills it for me. And usually

all I have to do is say make bindable and connect.

And there we go. But when I scroll back down, there it is. Now, Firecrawl is a shared resource. So, I can also manage

shared resource. So, I can also manage as the team administrator all of my organizations shared resources. There

they are. Here's the knowledge base, which is a SQL database. I have the firecrawl API. I can add other resources

firecrawl API. I can add other resources very easily, and it is just super easy to manage. Now, I also have an audit log

to manage. Now, I also have an audit log that shows me everything that happens within my team. I have a bunch of settings. I have some analytics. So, I

settings. I have some analytics. So, I

built out a good amount of the team functionality because that was the most exciting use case for me and my team.

There's a bunch of other features. I

spent a lot of time building out trust between agents and kits. So, for

example, when an agent is using a kit, if it runs into issues, it can actually tell the kits author about those issues.

an author or a publisher eventually gets a reputation score. If their kits are useful and valuable and safe, they can build up a reputation, which is super

nice. We also have community flagging

nice. We also have community flagging for spam and malicious content, although I also do my own scanning before things actually show up in the public registry.

And so, there's a lot more features I'm not going to get into. I want you to try it. This is my ask to you. I want you to

it. This is my ask to you. I want you to try it and tell me your feedback. I

don't care if it's constructive feedback, positive feedback, bugs that you find, ways that it could be better.

Drop it all in the comments of this video or reply down below if you're watching this on Twitter. My goal is to make it as easy for agents to discover

and install new end-to-end workflows. I

want every agent to become a thousand times more powerful. I also want to make it super easy for teams. And the other cool thing about using this inside of an enterprise organization or with your

team is that you can keep all the other agents in sync. You can make sure they're using the latest version or a specific version. You can make sure that

specific version. You can make sure that they have the right context, the right authentication. All of it stays in sync,

authentication. All of it stays in sync, easily manageable. You're not having

easily manageable. You're not having each agent build out their own version of something based on a prompt. That is

the key. If you enjoyed this video, please go check out Journey Kits.

journeykids.ai.