Agential AI with asynchronous function execution in a Redis queue. With this cognitive architecture, an AI agent can call dozens of parallel functions to solve the issue of hallucinations, source external information with each response, and massively speed up the time-to-first-token as compared to serial function execution.
“Intelligence is a fixed goal with variable means of achieving it.”
-William James
Designing cognitive architectures capable of agency is a notoriously difficult task. Base language models are confined to the knowledge of their training, and retrieval-augmented generation (RAG) pipelines generally follow a rigid pipeline for execution.
Redis-Agent leverages Chain-of-Thought reasoning to generate a plan of action for complex questions, and then offloads the task of function calling to a queue of Redis workers for concurrent execution.
The agent is hosted in a Flask server, and the execution of each function happens asynchronously first in first out from the queue of workers. Conversation history is also cached in Redis and summarized with 4o-mini, maintaining a fluid conversation while keeping within the LLM's context window. The whole architecture is deployable to your local machine with Docker and personal API keys, with a local NextJS client as the UI.
Free to use with an MIT license!
contact@attuneintelligence.com