Knowledge Quick Start

Get started with knowledge-based agents in 5 minutes

Knowledge Quick Start

Give your agent access to documents and get answers with citations.

Installation

pip install "praisonaiagents[knowledge]"

1. Simplest Usage (Default)

from praisonaiagents import Agent

agent = Agent(
    name="Research Assistant",
    knowledge=["research.pdf"]  # Just pass files
)

response = agent.start("What are the key findings?")

Where is my data stored? By default, the Chroma store persists at <project>/.praisonai/knowledge/chroma (absolute, per project). Override it with knowledge={"vector_store": {"config": {"path": "..."}}}. See Knowledge Storage.

2. Multiple Sources (Array)

from praisonaiagents import Agent

agent = Agent(
    name="Research Assistant",
    knowledge=[
        "paper.pdf",
        "notes.txt", 
        "docs/",           # Entire directory
    ]
)

response = agent.start("Summarize all documents")

Knowledge.add(url) — supported. As of PraisonAI PR #4334 (2026-08-25) the standalone Knowledge.add() fetches and ingests remote URLs via MarkItDown — the same pipeline used for local files.

from praisonaiagents import Knowledge

kb = Knowledge(config={"vector_store": {"provider": "chroma"}})
kb.add("https://example.com/article", user_id="user123")

Agent(knowledge=[url]) — still skipped. URL sources passed directly to Agent(knowledge=[...]) are not yet supported in the core SDK. Since PraisonAI PR #4004 any http:// or https:// entry is logged with a warning (scheme://hostname only, so tokens never leak) and skipped. This Agent-side code path lives in agent.py and is not touched by #4334. To ingest a URL through an agent, build a shared Knowledge instance, call kb.add(url), then pass knowledge=kb:

from praisonaiagents import Agent, Knowledge

kb = Knowledge(config={"vector_store": {"provider": "chroma"}})
kb.add("https://example.com/article")

agent = Agent(name="Reader", instructions="Summarise the article.", knowledge=kb)

3. Custom Configuration (Dict)

from praisonaiagents import Agent

agent = Agent(
    name="Research Assistant",
    knowledge={
        "sources": ["research.pdf", "data/"],
        "chunker": {
            "type": "semantic",      # Better topic boundaries
            "chunk_size": 512
        },
        "vector_store": {
            "provider": "chroma",
            "config": {"collection_name": "my_research"}
        }
    }
)

response = agent.start("What methodology was used?")

4. With Retrieval Options

from praisonaiagents import Agent

agent = Agent(
    name="Research Assistant",
    knowledge={
        "sources": ["research.pdf"],
        "retrieval_k": 5,            # Number of chunks
        "rerank": True               # Better relevance
    }
)

result = agent.query("What are the conclusions?")
print(result.answer)
print(result.citations)

5. Shared Knowledge (Instance)

from praisonaiagents import Agent, Knowledge

# Create shared knowledge base
kb = Knowledge(config={
    "vector_store": {"provider": "chroma"}
})
kb.add("company_docs/")

# Multiple agents share same knowledge
analyst = Agent(name="Analyst", knowledge=kb)
writer = Agent(name="Writer", knowledge=kb)

Supported File Types

Agent(knowledge=[...]) accepts local files, directories, and raw text. Knowledge.add(...) additionally accepts remote URLs.

Input	Accepted by Agent(knowledge=[...])	Accepted by Knowledge.add(...)
Documents (PDF, DOC, DOCX, TXT, MD, RTF)	Yes	Yes
Data (CSV, JSON, XML, Excel)	Yes	Yes
Directories	Yes — every supported file inside	Yes
Raw text strings	Yes — stored directly	Yes
Local HTML files (.html, .htm)	Yes	Yes
Remote URLs (http://, https://)	No — warned and skipped (PR #4004)	Yes — fetched via MarkItDown (PR #4334)

Next Steps

Storage Options

Configure vector stores

Chat with PDFs

Build a PDF chat agent

Chunking Strategies

Optimize document splitting

RAG

Advanced retrieval