This post is part of the technical series introducing Azara.
If you are keen to try out the dev features of Azara, you can sign up for the SDK release here
In the previous post of this technical series, we covered core components and how to build them. As Azara is a no-code platform for non-technical users, it is optimized to
Creating Agentic Scenarios
In order to make debugging of scenarios easier, we have been implementing a scenario editor and developer tools - early prototype seen below.
Langchain’s Langgraph Studio
However, Note that Langgraph Studio is now already available in Beta and is an excellent tool for just this purpose. Though it’s only available on MacOS currently.
https://github.com/langchain-ai/langgraph-studio preview below
Developer Tools: Plugin Creator
We have also created a plugin creator which is still WIP, but will allow non technical users to create plugins using a UI or by using the reverse engineering Agent. Some preview screenshots of this tool can be seen below.
The plugin creator is designed to provide a UI for those who aren't technical or who prefer an interface rather than a code editor to interact with..
First, you specify the logo, name and description for your plugin that will be shown to the user in the integrations tab.
Next, the user has several options to build out the functionality of the plugin
They can edit an existing plugin
They can extend the plugin by creating new routes
They can create the plugin automatically by using the AI Plugin generator
The latter is currently undergoing QA/QC - but it works by reverse engineering a library in github, and implementing it using an example plugin as a reference.
One of the main factors, why we were able to make this doable, is that the plugins themselves are very lean, and DRY implementations. Most of the coding implementation that needs to be done, is the UI user facing metadata.
When the user opts to create a method/route scaffold, they will see the following screen.
It's useful to note that for plugins, each route is in fact a langchain custom @tool, so you are in effect supplying the structured inputs and outputs for the method here.
Once the Methods scaffolds have been created using the above screen, or by importing a library and generating a plugin: You will be taken to the code editor where you can make any code changes or implement the methods in the scaffold.
You will be able to test the plugin at this point:
That it compiles without error, and
you can chat to the plugin using function calling from an Azara agent locally
SDK
We are currently in the process of refining our SDK. The sdk will be a subset of the Openapi.json created by the FastAPI server, with enhanced security and focused on the developer subset of functions.
Initially, we plan to release Python, Go, Typescript versions in Q4 2024.
For now - the most powerful functionality provided by the SDK will be the ability to extend the base platform webUI, Web Widgets (WW), plugins, and channels by directly uploading a plugin or agentic scenario.
These extend the platform in the following ways:
plugins - Integration plugins allow the platform to interact with external 3rd party services
channels - plugins that allow an agent to listen to a 3rd party service and respond, e.g. whatsapp, slack, email, telephony, etc.
agentic scenarios - where integrations plugins allow you to query or take action using an external service, scenarios allow you to implement different agent behaviour, e.g. Agentic RAG or custom multi-hop reasoning etc.
Next Steps
If you are interested in adopting Azara.ai for your projects or for your clients, please don’t hesitate to sign up to the waiting list below, and reach out to myself on X or by email at the contacts at the bottom of this blog post.
Do sign up to get notified when the above tools become available in beta, at the link here:
Thank you.
contacts:
steve@azara.ai aka madscientist
@steve_messina
@ai_azara
Commentaires