LongMemEval agent-memory evaluation
We tested several approaches to long-term agent memory on LongMemEval, including a combination of raw conversation retrieval and extracted facts.
Read the research article →
Redis AI Research
Research projects from the Redis AI team
This site collects our public experiments, benchmarks, and reference implementations. We share the code, results, and notes as we go.
Browse the repository →
Projects
Current work
Some projects answer a research question. Others are practical examples built to test an idea.
The Geometry of Agent Memory
We tested whether the geometry of retrieved memory embeddings can help identify questions an agent memory system is likely to answer incorrectly.
Read the research article →coding-agents
A FastAPI service for running coding agents in isolated workspaces, with authentication, streaming, and parallel tasks.
Open project →learning-agents
An agent that answers questions about CSV and JSON data by writing and running pandas code. Redis stores cached results and guidance from earlier errors.
Open project →memorydb-semantic-search
A benchmark of semantic search over LongMemEval chat sessions using AWS MemoryDB, S3, and EBS.
Open project →opencode-spec-optimization
An experiment that rewrites AGENTS.md, reruns the same OpenCode
questions, and keeps changes that improve the measured results.
agentic-ai-summit
Slides and source files from our Agentic AI Summit talk.
Open project →Areas we work on
- Semantic and vector search
- Caching for LLM applications
- Agentic systems and memory architectures
- RAG pipelines and evaluations
- Benchmarks for AI workloads on Redis