💾 Knowledge Base - DevDuck RAG Integration

🧠

Seamless Long-term Memory

DevDuck automatically retrieves relevant context before each query and stores conversations after. No manual RAG calls needed—just set an environment variable.

🔄 Automatic RAG Flow

When DEVDUCK_KNOWLEDGE_BASE_ID is set, DevDuck performs automatic retrieval and storage:

┌─────────────────────────────────────────────────────────────────────┐
│                        DevDuck Query Flow                          │
├─────────────────────────────────────────────────────────────────────┤
│                                                                     │
│   User Query                                                        │
│       │                                                             │
│       ▼                                                             │
│   ┌─────────────────────────────────────────┐                       │
│   │  1. RETRIEVE - Before Agent Runs           │                       │
│   │                                         │                       │
│   │  retrieve(                              │                       │
│   │    text=query,                          │                       │
│   │    knowledgeBaseId=KB_ID                │                       │
│   │  )                                      │                       │
│   │                                         │                       │
│   │  → Injects relevant context into agent   │                       │
│   └─────────────────────────────────────────┘                       │
│       │                                                             │
│       ▼                                                             │
│   ┌─────────────────────────────────────────┐                       │
│   │  2. AGENT EXECUTION                       │                       │
│   │                                         │                       │
│   │  Agent processes query with context     │                       │
│   │  from knowledge base                    │                       │
│   └─────────────────────────────────────────┘                       │
│       │                                                             │
│       ▼                                                             │
│   ┌─────────────────────────────────────────┐                       │
│   │  3. STORE - After Agent Completes          │                       │
│   │                                         │                       │
│   │  store_in_kb(                           │                       │
│   │    content="Input: ... Result: ...",    │                       │
│   │    title="DevDuck: 2026-02-02 | ...",   │                       │
│   │    knowledge_base_id=KB_ID              │                       │
│   │  )                                      │                       │
│   │                                         │                       │
│   │  → Saves for future reference            │                       │
│   └─────────────────────────────────────────┘                       │
│                                                                     │
└─────────────────────────────────────────────────────────────────────┘

⚡ Quick Start

🚀 One Environment Variable

Just set your Bedrock Knowledge Base ID and DevDuck handles the rest automatically:

# Enable automatic RAG
export DEVDUCK_KNOWLEDGE_BASE_ID="your-kb-id-here"

# Start DevDuck
devduck

# Every query now:
# 1. Retrieves relevant context from KB
# 2. Processes with enriched context
# 3. Stores conversation for future

Manual Tool Usage

# Retrieve from knowledge base
retrieve(
    text="how do I configure bedrock?",
    knowledgeBaseId="your-kb-id",
    numberOfResults=5,
    score=0.4  # Minimum relevance threshold
)

# Store content in knowledge base
store_in_kb(
    content="Important information to remember...",
    title="Session Notes - 2026-02-02",
    knowledge_base_id="your-kb-id"
)

🔍 Retrieve Tool

Semantic search against your Bedrock Knowledge Base with relevance scoring.

Parameter	Default	Description
`text`	required	Query text for semantic search
`knowledgeBaseId`	env var	Bedrock Knowledge Base ID
`numberOfResults`	5	Maximum results to return
`score`	0.4	Minimum relevance score (0.0-1.0)
`region`	us-west-2	AWS region
`enableMetadata`	false	Include source metadata in results

Response Format

{
  "content": {
    "text": "Retrieved document content...",
    "type": "TEXT"
  },
  "location": {
    "customDocumentLocation": {
      "id": "document_id"
    },
    "type": "CUSTOM"
  },
  "metadata": {
    "x-amz-bedrock-kb-source-uri": "source_uri",
    "x-amz-bedrock-kb-chunk-id": "chunk_id",
    "x-amz-bedrock-kb-data-source-id": "data_source_id"
  },
  "score": 0.95
}

💾 Store Tool

Real-time ingestion into Bedrock Knowledge Base. Runs async in background thread.

Parameter	Default	Description
`content`	required	Text content to store
`title`	timestamp	Optional title for the content
`knowledge_base_id`	env var	Target Knowledge Base ID

# Store with custom title
store_in_kb(
    content="Complete API documentation for the user service...",
    title="API Docs - User Service v2.0",
    knowledge_base_id="kb-12345"
)

# Store with auto-generated timestamp title
store_in_kb(content="Important meeting notes...")

✨ Features

🔄

Automatic RAG

Retrieval and storage happen automatically—no manual tool calls needed.

🧠

Cross-Session Memory

Conversations persist across sessions. DevDuck remembers past interactions.

⚡

Async Storage

Storage runs in background thread. No impact on response latency.

📊

Relevance Scoring

Filter results by relevance score. Only high-quality context injected.

🔍

Semantic Search

Vector-based similarity matching. Finds conceptually related content.

🏷️

Metadata Tracking

Track sources, chunk IDs, and data sources for audit trails.

⚙️ Configuration

Variable	Description
`DEVDUCK_KNOWLEDGE_BASE_ID`	Bedrock Knowledge Base ID for automatic RAG
`STRANDS_KNOWLEDGE_BASE_ID`	Alternative env var for KB ID (used by strands_tools)
`AWS_REGION`	AWS region for Bedrock (default: us-west-2)

💡 Use Cases

📚

Documentation Assistant

Index project docs for instant retrieval during development.

🔄

Session Continuity

Pick up where you left off across multiple DevDuck sessions.

👥

Team Knowledge

Share knowledge across team members through shared KB.

🧪

Research Memory

Build up knowledge over long research projects.