Can we use chat history in workflow?

[alaap001]

Hey,

A simple question: I tried using read_chat_history=True in a workflow and when in the playground I told Agent my name, and in the next message I asked it what was my name butt it couldn't answer.
Replied: "I'm sorry, I don't retain conversation-specific context between interactions. How can I assist you today?"

Is there any way we can maintain chat/message history for workflow sessions?

Since a user can have a complex run method with multiple agents communicating we define the sequence they work on, something like: #1612 (comment)

But then we also want it to work as a chatbot, but with complex logic in place. How can we do that in a workflow? Knowing chat history is important as well for chatbot to feel natural and work properly.

Appreciate any help.
Thanks :)

[dirkbrnd]

Are you sure you are storing the chat history? Are you passing your request to the team leader?

If you can supply a code snippet we can check.

[alaap001]

Sure so I'll explain in detail.
Let's say we have a simple workflow as below:

from pydantic import BaseModel, Field
from phi.agent import Agent
from phi.model.openai import OpenAIChat
from phi.workflow.workflow import Workflow
from phi.tools.postgres import PostgresTools
from phi.run.response import RunResponse
from phi.utils.pprint import pprint_run_response
from phi.model.ollama import Ollama
from dotenv import load_dotenv
from phi.embedder.ollama import OllamaEmbedder
from phi.document.chunking.agentic import AgenticChunking
from marker_chunking import MarkerChunking
import re
from phi.storage.agent.postgres import PgAgentStorage


load_dotenv()

class AnalysisWorkflow(Workflow):
    description: str = "SQL Analysis Workflow"
    db_url: str = "postgresql+psycopg://ai:ai@localhost:5432/ai"

    intent_recog: Agent = Agent(
        name="Smart Assistant",
        description="""You are SmartAssistant.""",
        instructions=[
            "You will understand the user question and understand the intent behind the question",
            "then give the intent"
            ],
            read_chat_history=True,
            # add_datetime_to_instructions=True,
            # storage=PgAgentStorage(table_name="agent_sessions", db_url=db_url),
            # add_history_to_messages=True,
            # add_name_to_instructions=True,
            # model=Ollama(id="qwen2.5:32b-instruct-q8_0"),
            model=Ollama(id="deepseek-r1:32b-qwen-distill-q8_0"),
            # model=Ollama(id="phi4:14b-q8_0"),
            # model=OpenAIChat(id="gpt-4o-mini", temperature=0.1),
            # response_model=IntentRecog,
            # structured_outputs=True,
            markdown=True
    )

    def run(self, question: str) -> RunResponse:

        intent_recog_result = self.intent_recog.run(question)
        
        return RunResponse(content=intent_recog_result.content) 


from phi.playground import Playground, serve_playground_app
from phi.storage.workflow.sqlite import SqlWorkflowStorage
from phi.tools.postgres import PostgresTools

analysis = AnalysisWorkflow(
    debug_mode=True,
    workflow_id="analysis",
)

app = Playground(
    workflows=[analysis]
).get_app()

if __name__ == "__main__":
    serve_playground_app("playgroundHistCheck:app", reload=True)

If I run this with read_chat_history=True

below is the ourput:

if instead I use:

storage=PgAgentStorage(table_name="agent_sessions", db_url=db_url),
add_history_to_messages=True,```

Then it works.

so 2 questions here.

What is the difference btw the 2? and shouldn't read chat history work in the workflow playground as well? since it works in normal chat.

Also read_chat_history uses a tool making it a limitation because now I cannot use small and good models like phi4 14b with read_chat_history because it doesn't support tools. So that limits many open source models from using chat history.

Thanks for the quick assistance.