GenerativeAI is not the only way to use AI. I personally feel Wappler should look at AI from the perspective of helping users to build AI applications, rather than considering AI only for the generative/code aspect. I did message the team re: this, but didn't get a response.
Interesting, can you give an example? I can't think of a way that AI would help any of my applications. But I am very interested in your thoughts.
Every app / product is different. For example, my current project has a lot of multi-step AI workflows. I plan to integrate Langchain to build an agentic application myself, but the team would do a better job of this and I think Wappler would be well suited to this imo.
From ChatGPT:
To integrate LangChain natively into Wappler, you’d essentially want to leverage Wappler’s low-code capabilities alongside LangChain’s agentic framework. The key is to allow Wappler users to create AI-powered applications without writing extensive custom code, but with the flexibility to customize AI agents and workflows when needed. Here’s a breakdown of how you could implement this:
To integrate LangChain natively into Wappler, you’d essentially want to leverage Wappler’s low-code capabilities alongside LangChain’s agentic framework. The key is to allow Wappler users to create AI-powered applications without writing extensive custom code, but with the flexibility to customize AI agents and workflows when needed. Here’s a breakdown of how you could implement this:
1. LangChain as a Wappler Custom Module
Wappler allows the creation of custom modules, which is how you could encapsulate LangChain’s functionality for end users. A custom module would expose key LangChain components to the Wappler interface, allowing users to configure their AI agents visually.
• Create a LangChain SDK Integration: Develop a custom Wappler module that directly interacts with LangChain’s SDK. This module would expose the LangChain components like Chains, Agents, Tools, and Prompts as configurable inputs in Wappler.
• JSON Representation of Chain Configurations: Wappler primarily works with JSON-based configurations, so you can allow users to define chains, agents, and tools in a JSON format behind the scenes, while giving them UI components to set up these configurations.
2. Abstract LangChain Components into Wappler UI
You’ll want to make LangChain’s complexity abstracted for the end users, similar to how Wappler handles other low-code functionalities:
• Agent Configuration Interface: Create a visual configuration interface for LangChain agents. For example, users could select from a dropdown of predefined tools or models (LLMs), or create custom prompts. This would generate the corresponding LangChain setup behind the scenes.
• Let users define agent behavior, input parameters, and tools it can call (e.g., search, calculation APIs).
• For more advanced users, allow them to toggle into an advanced mode to edit the raw JSON output of the chain configuration.
• Dynamic Prompts and Chains: Allow users to define chains of reasoning or workflows visually. Each step in the chain could be represented as a block in the Wappler UI, which users can configure by dragging and dropping or adjusting parameters. This block can represent LangChain’s sequential or agentic logic.
• Add a visual editor where users can set prompts or define conditions for how the chain progresses.
3. Enable Custom Logic and External Tools
LangChain’s strength lies in its ability to call external APIs and tools. Integrating this into Wappler would mean enabling users to extend their agentic systems without leaving the platform:
• Connect External APIs: Provide a Wappler interface where users can link external APIs as tools for the agents to call. For example, a user might set up an API connection to their internal database or a third-party service. You can then allow them to use LangChain agents to query that API dynamically as part of the workflow.
• Custom Code Hooks: In cases where Wappler users want more control, offer a way to insert custom code (JavaScript, Python) that LangChain agents can call. This would allow more advanced users to extend the agent’s behavior while keeping the overall workflow visual.
4. Store and Manage State in Wappler
Since LangChain can involve complex workflows that might require storing intermediate states or results, integrating LangChain into Wappler would need to ensure that state management is handled effectively.
• Use Wappler’s Built-in Database and Storage: Allow LangChain agents to store and retrieve context or information via Wappler’s existing database management tools. This could be done by automatically generating tables for storing agent interactions or by allowing users to define how results should be stored.
• Session-Based or Long-Term Memory: For agent systems requiring memory (e.g., remembering past interactions), build session storage capabilities where each agent interaction gets stored in a Wappler-accessible session or database.
5. Leverage LangChain’s API for Wappler Apps
LangChain’s models and tools can be accessed through APIs, so a native Wappler implementation could take advantage of LangChain’s hosted services:
• LLM API Integration: Let users select and integrate LLMs via API keys (OpenAI, Cohere, etc.) directly within Wappler. This would let them leverage the power of LangChain without the complexity of managing infrastructure.
• Chain Execution via API: Provide an interface where users can create workflows or chain executions in LangChain and trigger them from their Wappler app via an API call.
etc..
There are already some great AI sdk integrations as custom extensions from @Hyperbytes like:
Maybe somebody can build one for LangChain
Have taken a quick look at Langchain, has an API and there is also an npm module so probably possible to do it as a custom module.
I really need to spend some time looking at what benefits it will bring over ChatGPT, Gemini and Claude..
There are lots of new AI platforms appearing but many are just interfaces to the above services and i need to be convinced of the extra value they will bring
Langchain isn’t an LLM so it’s not comparable to chatgpt etc. It’s for building agentic applications. Eg chaining together LLMs of your choice.