# Context-aware retrieval for AI Agents

Stop stuffing context windows. Power your agents with a stateful relevance engine that retrieves chunks based on user history and intent, not just semantic similarity.

## Faster retrieval, smarter context

### Build in-house

#### Retrieval logic
- **Similarity Only**
- **Relevance (Similarity + User State)**

| Feature           | Current Approach                    | With Shaped                     |
|-------------------|-------------------------------------|---------------------------------|
| Context window    | Stuffed with random chunks          | Optimized with high-precision chunks |
| User memory       | Manual lookups in Redis             | First-class input in the query  |
| Latency           | High (Network hops)                 | <50ms (Vertically integrated)   |

### Give your agents memory this week

#### Day 1
1. **Connect Data**  
   Ingest Knowledge Base & User Events streams.

#### Days 2-6
2. **Configure Logic**  
   Define ranking signals in declarative YAML.

#### Day 7
7. **Deploy agent**  
   Push to production with <50ms latency.

### Reduce token costs
Stop wasting tokens on irrelevant chunks. Filter retrieval by user intent before hitting the LLM context window.

### Long-term memory
Standard vector DBs are amnesiac. Shaped natively stores user interactions, giving your agent instant access to past behavior.

## Ship stateful agents in one sprint

### Step 1: Connect your data
Ingest your documents, vector embeddings, and—crucially—user interaction streams into a single unified schema. No complex ETL required.

### Step 2: Configure retrieval logic
Stop writing Python glue code to filter chunks. Define complex retrieval strategies—combining vector similarity, keyword matching, and user history—in a single declarative query.

### Step 3: Deploy automatically
Shaped handles the infrastructure—training, scheduling, and auto-scaling your retrieval endpoint. Ship production-ready agent memory without managing vector indices or inference servers.

## One context engine, every agent
Whether you are building a customer support bot, a shopping assistant, or an internal research tool, they all share the same unified understanding of the user.

## Solutions for every platform

**Semantic search**  
Retrieve knowledge based on meaning using modern embedding models.

**User history**  
Personalize responses based on past clicks, views, and chat sessions.

**Hybrid retrieval**  
Combine keyword matching (BM25) with vector search for precise recall.

**Metadata filtering**  
Apply hard constraints (availability, permissions) within the retrieval query.

# Ready to fix your RAG pipeline?
Join engineering teams using Shaped to drive engagement and revenue.

100M+

Vectors indexed

<50ms

p99 latency

1B+

Predictions