Unable to correctly use knowledge. Any thoughts?

[jtbwatson]

Trying to build a simple flow that has some Markdown files of network configuration. Running into an issue where, regardless of what I prompt with or ask, I always just get a summary of the documents.

Too long to post the responses here, but this is my current setup:

from praisonaiagents import Agent, Task, PraisonAIAgents
import os
from dotenv import load_dotenv

load_dotenv()

config = {
    "vector_store": {
        "provider": "chroma",
        "config": {"collection_name": "praison", "path": ".praison"},
    },
    "llm": {
        "provider": "ollama",
        "config": {
            "ollama_base_url": os.getenv("OLLAMA_HOST"),
            "model": os.getenv("OLLAMA_MODEL"),
            "temperature": 0.7,
            "max_tokens": 4096,
        },
    },
    "embedder": {
        "provider": "ollama",
        "config": {
            "model": os.getenv("OLLAMA_EMBEDDING_MODEL"),
            "ollama_base_url": os.getenv("OLLAMA_HOST"),
            "embedding_dims": 1536,
        },
    },
}

knowledge_agent = Agent(
    name="KnowledgeAgent",
    role="Information Specialist",
    goal="Store and retrieve knowledge efficiently",
    backstory="Expert in managing and utilizing stored knowledge",
    knowledge=["network_docs/wireless_infrastructure.md",
              "network_docs/datacenter_infrastructure.md",
              "network_docs/internet_traffic_flow.md",
              "network_docs/wan_traffic_flow.md",
              "network_docs/wired_campus_infrastructure.md"],
    knowledge_config=config,
    verbose=True,
)

# Define a task for the agent
knowledge_task = Task(
    name="knowledge_task",
    description="How are the core switches setup?",
    expected_output="Answer to the question",
    agent=knowledge_agent,
)

# Create and start the agents
agents = PraisonAIAgents(
    agents=[knowledge_agent],
    tasks=[knowledge_task],
    process="sequential",
    user_id="user1",
)

# Start execution
result = agents.start()

No matter what I ask it, whether it's about network or if I ask it how to make peach cobbler, it always begins just summarizing the documents that have been passed through. I can't figure out a way to make it just reference the documents when it needs them or pull out bits and pieces of what I'm asking it for.

Any pointers would be greatly appreciated.

[jtbwatson]

It seems the issue is that the way PraisonAI is approaching this is by loading the whole document into the prompt.. I have a custom app I have built that I'm going to end up using instead until this is worked out more effectively. Unfortunate that this isn't performing as anticipated.