Agents with Knowledge

Knowledge stores domain-specific content that can be added to the context of the agent to enable better decision making.

The Agent can search this knowledge at runtime to make better decisions and provide more accurate responses. This searching on demand pattern is called Agentic RAG.

Knowledge for Agents

Kern Agents use Agentic RAG by default, meaning when we provide knowledge to an Agent, it will search this knowledge base, at runtime, for the specific information it needs to achieve its task.

For example:

1import asyncio
2
3from kern.agent import Agent
4from kern.db.postgres.postgres import PostgresDb
5from kern.knowledge.embedder.openai import OpenAIEmbedder
6from kern.knowledge.knowledge import Knowledge
7from kern.vectordb.pgvector import PgVector
8
9db = PostgresDb(
10    db_url="postgresql+psycopg://ai:ai@localhost:5532/ai",
11    knowledge_table="knowledge_contents",
12)
13
14# Create Knowledge Instance
15knowledge = Knowledge(
16    name="Basic SDK Knowledge Base",
17    description="Kern 2.0 Knowledge Implementation",
18    contents_db=db,
19    vector_db=PgVector(
20        table_name="vectors",
21        db_url="postgresql+psycopg://ai:ai@localhost:5532/ai",
22        embedder=OpenAIEmbedder(),
23    ),
24)
25# Add from URL to the knowledge base
26asyncio.run(
27    knowledge.ainsert(
28        name="Recipes",
29        url="https://kern-public.s3.amazonaws.com/recipes/ThaiRecipes.pdf",
30        metadata={"user_tag": "Recipes from website"},
31    )
32)
33
34agent = Agent(
35    name="My Agent",
36    description="Kern 2.0 Agent Implementation",
37    knowledge=knowledge,
38    search_knowledge=True,
39)
40
41agent.print_response(
42    "How do I make chicken and galangal in coconut milk soup?",
43    markdown=True,
44)

We can give our agent access to the knowledge base in the following ways:

• We can set search_knowledge=True to add a search_knowledge_base() tool to the Agent. search_knowledge is True by default if you add knowledge to an Agent.
• We can set add_knowledge_to_context=True to automatically add references from the knowledge base to the Agent's context, based in your user message. This is the traditional RAG approach.

Custom knowledge retrieval

If you need complete control over the knowledge base search, you can pass your own knowledge_retriever function with the following signature:

1def knowledge_retriever(agent: Agent, query: str, num_documents: Optional[int], **kwargs) -> Optional[list[dict]]:
2  ...

Example of how to configure an agent with a custom retriever:

1def knowledge_retriever(agent: Agent, query: str, num_documents: Optional[int], **kwargs) -> Optional[list[dict]]:
2  ...
3
4agent = Agent(
5    knowledge_retriever=knowledge_retriever,
6    search_knowledge=True,
7)

This function is called during search_knowledge_base() and is used by the Agent to retrieve references from the knowledge base.

Knowledge storage

Knowledge content is tracked in a "Contents DB" and vectorized and stored in a "Vector DB".

Contents database

The Contents DB is a database that stores the name, description, metadata and other information for any content you add to the knowledge base.

Below is the schema for the Contents DB:

This data is best displayed on the knowledge page of the AgentOS UI.

Vector databases

Vector databases offer the best solution for retrieving relevant results from dense information quickly.

Adding contents

The typical way content is processed when being added to the knowledge base is:

For example, to add a PDF to the knowledge base:

1...
2knowledge = Knowledge(
3    name="Basic SDK Knowledge Base",
4    description="Kern 2.0 Knowledge Implementation",
5    vector_db=vector_db,
6    contents_db=contents_db,
7)
8
9asyncio.run(
10    knowledge.ainsert(
11        name="CV",
12        path="cookbook/08_knowledge/testing_resources/cv_1.pdf",
13        metadata={"user_tag": "Engineering Candidates"},
14    )
15)

Example: Agentic RAG Agent

Let's build a RAG Agent that answers questions from a PDF.

1docker run -d \
2  -e POSTGRES_DB=ai \
3  -e POSTGRES_USER=ai \
4  -e POSTGRES_PASSWORD=ai \
5  -e PGDATA=/var/lib/postgresql/data/pgdata \
6  -v pgvolume:/var/lib/postgresql/data \
7  -p 5532:5432 \
8  --name pgvector \
9  agnohq/pgvector:16

1uv pip install -U pgvector pypdf psycopg sqlalchemy

1uv pip install -U pgvector pypdf psycopg sqlalchemy

1import asyncio
2from kern.agent import Agent
3from kern.models.openai import OpenAIResponses
4from kern.knowledge.embedder.openai import OpenAIEmbedder
5from kern.knowledge.knowledge import Knowledge
6from kern.vectordb.pgvector import PgVector
7
8db_url = "postgresql+psycopg://ai:ai@localhost:5532/ai"
9
10db = PostgresDb(
11    db_url=db_url,
12    knowledge_table="knowledge_contents",
13)
14
15knowledge = Knowledge(
16    contents_db=db,
17    vector_db=PgVector(
18        table_name="recipes",
19        db_url=db_url,
20        embedder=OpenAIEmbedder(),
21    )
22)
23
24agent = Agent(
25    model=OpenAIResponses(id="gpt-5.2"),
26    db=db,
27    knowledge=knowledge,
28    markdown=True,
29)
30if __name__ == "__main__":
31    asyncio.run(
32        knowledge.ainsert(
33            name="Recipes",
34            url="https://kern-public.s3.amazonaws.com/recipes/ThaiRecipes.pdf",
35            metadata={"user_tag": "Recipes from website"}
36        )
37    )
38    # Create and use the agent
39    asyncio.run(
40        agent.aprint_response(
41            "How do I make chicken and galangal in coconut milk soup?",
42            markdown=True,
43        )
44    )

1python agentic_rag.py

Developer Resources

• View the Agent schema
• View the Knowledge schema
• View Cookbook