https://github.com/redis/redis-ai-research-public
Home of AI Research at Redis
https://github.com/redis/redis-ai-research-public
Last synced: about 1 month ago
JSON representation
Home of AI Research at Redis
- Host: GitHub
- URL: https://github.com/redis/redis-ai-research-public
- Owner: redis
- License: mit
- Created: 2026-04-29T09:28:22.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2026-06-30T17:10:20.000Z (about 1 month ago)
- Last Synced: 2026-06-30T19:09:41.028Z (about 1 month ago)
- Language: Python
- Homepage:
- Size: 1.72 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Redis AI Research
Public-facing projects, experiments, and reference implementations from the **Redis AI Research team**.
This monorepo collects open-source work exploring how Redis can be applied to modern AI workloads — semantic caching, vector search, retrieval-augmented generation (RAG), agentic systems, and beyond.
## What you'll find here
Each top-level directory is a self-contained project with its own README, dependencies, and setup instructions. Projects generally fall into one of these areas:
- **Semantic & vector search** — using Redis as a vector store for embeddings, similarity search, and hybrid retrieval.
- **Caching for LLM applications** — semantic caching, result reuse, and latency optimization patterns.
- **Agentic systems** — multi-agent orchestration, learning agents, and memory architectures backed by Redis.
- **RAG pipelines** — reference implementations and benchmarks for retrieval-augmented generation.
- **Benchmarks & evaluations** — performance studies and methodology for AI workloads on Redis.
## Repository layout
```
.
├── coding-agents/ # multi-tenant OpenAI Agents SDK coding-agent
│ # service with sandboxed tenant workspaces
├── learning-agents/ # multi-agent NL → pandas analysis with Redis-backed
│ # semantic cache and learned-guidance memory
├── memorydb-semantic-search/ # vector search over ~238k LongMemEval sessions on
│ # AWS MemoryDB (pure-Redis vs. hybrid+S3 vs. hybrid+EBS)
├── opencode-spec-optimization/ # iteratively rewrites AGENTS.md until OpenCode
│ # runs meet per-objective success thresholds
└── ...
```
Projects are added incrementally. Browse the top-level directories to see what's currently available.
### Current projects
- **[coding-agents](coding-agents/)** — A multi-tenant FastAPI coding-agent service built on the OpenAI Agents SDK sandbox harness. Authenticates tenants, maps each request to an isolated workspace, supports synchronous and SSE-streamed task execution, and can fan work out to parallel sandbox subagents.
- **[learning-agents](learning-agents/)** — A multi-agent system that turns natural-language questions into executable pandas code over a CSV/JSON dataset. Uses Redis vector indices for a semantic result cache and a persistent "guidance" memory that learns from past errors to improve future retries.
- **[memorydb-semantic-search](memorydb-semantic-search/)** — Semantic search over ~238k LongMemEval chat sessions on AWS MemoryDB (Valkey 7.2 with search). Benchmarks three backend approaches — pure Redis, hybrid+S3, and hybrid+EBS — with Terraform for the full AWS stack (MemoryDB cluster, bastion, S3, VPC endpoint) and a FastAPI server exposing the search APIs.
- **[opencode-spec-optimization](opencode-spec-optimization/)** — Workflow that runs every question in `QUESTIONS.yaml` against [OpenCode](https://opencode.ai) multiple times in parallel, evaluates the responses against the thresholds in `OBJECTIVE.yaml`, and iteratively rewrites `AGENTS.md` until every objective passes or the retry limit is hit. Useful for tuning agent system prompts via measurable, repeatable evals.
## Getting started
1. Clone the repository:
```bash
git clone https://github.com/redis/redis-ai-research-public.git
cd redis-ai-research-public
```
2. Pick a project directory and follow its own README for setup and usage. Most projects assume a running Redis instance (Redis Stack or Redis 8+ with the search/vector modules).
## Prerequisites (common)
- Redis Stack or Redis 8+ (for vector search and JSON support)
- Python 3.11+ (most projects)
- An API key for the LLM provider used by the project (typically OpenAI, Anthropic, or Google)
Check each project's README for exact requirements.
## Contributing
Contributions, issues, and discussion are welcome. Open an issue to propose a new experiment, report a bug, or ask a question. Pull requests should target the relevant project directory and include updates to that project's README where appropriate.
## License
This repository is released under the [MIT License](LICENSE). See the `LICENSE` file for details.
## Contact
For questions about the AI Research team's work, open an issue in this repository.