NuRedis Hub · Architecture

The Model-Agnostic Consistency Bridge.

NuRedis is an MCP-served bridge over Redis Stack, RedisVL, and the Agent Memory Server. It turns Redis into a deterministic intelligence layer — enforcing consistency across any model (Gemini, OpenAI, Claude, Llama) through unified vector RAG, semantic caching, and real-time observability.

v0.1.0 Try it acecalisto3/NuRedis Repo github.com/acecalisto3/TryNuRedis Stack Python · Gemini · Redis Stack · Gradio

Topology

Infinite Models. One Source of Truth.

NuRedis decouples the 'brain' from the 'memory'. Whether you're routing through Google Gemini, hosted OpenAI, or local Ollama, the agent's context and consistency are governed by the NuRedis bridge, ensuring deterministic outputs regardless of provider variance. 

        ┌────────────────────────────────────────────────────────────────────┐
        │   NuRedis Chat UI (Gradio) — Multi-model playground                │
        │     select: Gemini | OpenAI | Claude | Ollama | Deterministic       │
        └─────────────────────────────────────┬──────────────────────────────┘
                                              │ 
                                              ▼
        ┌────────────────────────────────────────────────────────────────────┐
        │                       NuRedis Consistency Bridge                  │
        │   1. Semantic Cache → 2. Vector Recall → 3. Model Reason          │
        │   4. Routing Trace  → 5. Memory Remember                          │
        └─┬──────────┬──────────┬──────────┬──────────────────────────────────┘
          │          │          │          │
          ▼          ▼          ▼          ▼
    ┌─────────┐ ┌─────────┐ ┌────────┐ ┌────────┐
    │ Redis   │ │ RedisVL │ │  AMS   │ │Marimo  │
    │ Stack   │ │ vectors │ │ memory │ │ Observer │
    └─────────┘ └─────────┘ └────────┘ └────────┘

The Routing Trace

Visualizing Provenance.

Consistency is not a black box. NuRedis explicitly tracks the origin of every answer. Every agent turn is stamped with its 'Routing Trace', showing the exact path taken to reach the answer.

⚡
Cache Hit
Instant sub-10ms response from the Semantic Cache. Behavior is frozen and deterministic.
📚
RAG Grounded
Answer is 'anchored' in retrieved context from RedisVL. Model reasoning is grounded in your data.
🤖
LLM Direct
Answer from model weights. Used when context is weak. Labeled as 'Un-anchored' in the UI.
✅
Consistent
The final badge. Confirms the answer matches the source of truth with a high similarity score.

Provider Matrix

First-class Gemini, Cloud-ready.

NuRedis is optimized for the modern cloud ecosystem. While local-first at its core, it provides native high-performance bridges to Google's Gemini fleet, ensuring the best reasoning-to-cost ratio for public spaces.

Hosted
gemini
Primary cloud choice. Gemini 1.5 Flash + Google Text Embeddings.
Hosted
openai
Standard hosted choice. GPT-4o-mini + Text-Embedding-3-Small.
Local
ollama
Zero-network fallback. llama3.2 + nomic-embed-text.
Local
deterministic
The control. Offline SHA256-based provider for logic validation.

Observability

Live Matrix Monitoring.

Gradio Chat UI

The 'Front Door' for agents and users. Select providers, toggle RAG, and watch the 'Routing Trace' in real-time. Built-in support for Hugging Face Spaces with zero-config deployment.

python3 app.py · Available on Hugging Face at acecalisto3/TryNuRedis.

Marimo Observer

The 'Cockpit'. A reactive dashboard at localhost:2718 that renders the internal state of the bridge: stream throughput, health telemetry, and the vector index contents.

marimo run marimo/visualizer.py · Unified environment configuration for Gemini, OpenAI, and Redis keys.