During our open beta we are assessing the scalability and composability of PRISM. While we aim to deliver a consistent experience regardless, please understand that some features are still works in progress. If you discover any bugs, please reach out to our team in Discord!
What is PRISM?
PRISM is an AI/ML platform that allows businesses and developers to create and manage AI and ML tooling for blockchains. PRISM removes the heavy lifting of creating enterprise-level predictive and generative AI features into blockchain applications. PRISM empowers developers to abstract complexities away for end users by integrating custom-trained predictive analysis models, and state-of-the-art (SOTA) AI agent integrations.
Key Features
AI Agents
Users can create AI Agents in PRISM with a few simple starting inputs. PRISM Agents are built with propretiary language graphs (DAG) designed to assist with blockchain interagation and analysis. They are not simply wrappers around an LLM model! A single request to a PRISM Agent makes several LLM calls, and uses the DAG to determine optimal request routing.
Solana Centric
PRISM is optimized for the Solana ecosystem. Our toolkits are optimized to ensure that Agents can effectively interact and reason with real-time Solana data. We have partnered with several ecosystem leaders to ensure that the highest quality information is readily available to PRISM Agents.
Real-time Toolkits
LLM's are powerful because of the large corpus of information they are trained on. While SOTA techniques to reduce the time for finetuning a model can drastically reduce the time and expense to add new data, it's not feasible to retrain a new model every 400ms (the speed of block creation on Solana). PRISM Agents have access to an ever growing collection of toolkits that allow them to request additional data at inference time based on the input.
Custom Contexts
Custom Contexts are very alpha! Please report any issues you find in Discord!
While tools allow an agent to access real-time data, a user may want to provide a large amount of predefined data to their agent. Most modern LLM's provide Retrieval Augmented Generation (RAG) capabilities with a vectorstore. PRISM provides vector stores in the form of Contexts.
A user may schedule various types of sources to be loaded into their contexts, reducing the burden of manually formatting these properly for embedding and loading into the vectorstore.