I want to use the local LLM via Ollama, but I'm facing some issues. It keeps asking me for the OpenAI API key

Saeed_Kasmani · February 3, 2025, 7:43am

from crewai import Agent, Crew, Process, Task, LLM
from crewai.project import CrewBase, agent, crew, task
from crewai_tools import SerperDevTool, ScrapeWebsiteTool
from crewai.knowledge.source.string_knowledge_source import StringKnowledgeSource
from models import JobRequirements, ResumeOptimization, CompanyResearch
from langchain_ollama import OllamaEmbeddings

import os

# 🚀 Explicitly remove OpenAI API Key
os.environ.pop("OPENAI_API_KEY", None)
os.environ["CREWAI_LLM_PROVIDER"] = "ollama"
os.environ["CREWAI_EMBEDDINGS_PROVIDER"] = "ollama"

print("\n🔍 DEBUG: Checking CrewAI Environment Variables")
for key in ["OPENAI_API_KEY", "CREWAI_LLM_PROVIDER", "CREWAI_EMBEDDINGS_PROVIDER"]:
    print(f"{key}: {os.environ.get(key)}")

# Use Ollama embeddings
ollama_embeddings = OllamaEmbeddings(model="nomic-embed-text:latest")

# Initialize Ollama LLM
llm = LLM(
    model="ollama/deepseek-r1:32b",
    base_url="http://localhost:11434",
    api_key="ollama",  # This ensures it's not using OpenAI
    temperature=0.3
)
import json

print("\n🔍 DEBUG: CrewAI LLM Configuration:")
# print(json.dumps(llm.dict(), indent=4))

print("\n🔍 DEBUG: Environment Variables (Ensure No OpenAI Keys)")
for key in ["OPENAI_API_KEY", "CREWAI_LLM_PROVIDER", "CREWAI_EMBEDDINGS_PROVIDER"]:
    print(f"{key}: {os.environ.get(key)}")
# Initialize the tools
search_tool = SerperDevTool()
scrape_tool = ScrapeWebsiteTool()

@CrewBase
class ResumeCrew:
    """ResumeCrew for resume optimization and interview preparation"""

    def __init__(self, resume_text: str) -> None:
        self.resume_text = resume_text
        self.llm = llm
        self.agents_config = 'config/agents.yaml'
        self.tasks_config = 'config/tasks.yaml'

    @agent
    def resume_analyzer(self) -> Agent:
        return Agent(
            config=self.agents_config['resume_analyzer'],
            verbose=True,
            llm=self.llm,
            knowledge_sources=[StringKnowledgeSource(content=self.resume_text)]
        )

    @agent
    def job_analyzer(self) -> Agent:
        return Agent(
            config=self.agents_config['job_analyzer'],
            verbose=True,
            tools=[scrape_tool],
            llm=self.llm
        )

    @agent
    def company_researcher(self) -> Agent:
        return Agent(
            config=self.agents_config['company_researcher'],
            verbose=True,
            tools=[search_tool],
            llm=self.llm,
            knowledge_sources=[StringKnowledgeSource(content=self.resume_text)]
        )

    @agent
    def resume_writer(self) -> Agent:
        return Agent(
            config=self.agents_config['resume_writer'],
            verbose=True,
            llm=self.llm
        )

    @agent
    def report_generator(self) -> Agent:
        return Agent(
            config=self.agents_config['report_generator'],
            verbose=True,
            llm=self.llm
        )

    @task
    def analyze_job_task(self) -> Task:
        return Task(
            config=self.tasks_config['analyze_job_task'],
            output_file='output/job_analysis.json',
            output_pydantic=JobRequirements
        )

    @task
    def optimize_resume_task(self) -> Task:
        return Task(
            config=self.tasks_config['optimize_resume_task'],
            output_file='output/resume_optimization.json',
            output_pydantic=ResumeOptimization
        )

    @task
    def research_company_task(self) -> Task:
        return Task(
            config=self.tasks_config['research_company_task'],
            output_file='output/company_research.json',
            output_pydantic=CompanyResearch
        )

    @task
    def generate_resume_task(self) -> Task:
        return Task(
            config=self.tasks_config['generate_resume_task'],
            output_file='output/optimized_resume.md'
        )

    @task
    def generate_report_task(self) -> Task:
        return Task(
            config=self.tasks_config['generate_report_task'],
            output_file='output/final_report.md'
        )

    @crew
    def crew(self) -> Crew:
        return Crew(
            agents=self.agents,
            tasks=self.tasks,
            verbose=True,
            process=Process.sequential,
            knowledge_sources=[StringKnowledgeSource(content=self.resume_text)],
            cache=False,
            planning=True,
            planning_llm=self.llm,
            memory=False,
            embedder={
                "provider": "ollama",  # Ensure this is explicitly set
                "config": {
                    "model": "nomic-embed-text"
                }
            }
        )

eacee · February 4, 2025, 2:20am

It was my understanding that if you use the os.environ notation, you need to use a .env file. Also, it looks like you are missing load_dotemv().

This syntax is working for me:

# Load environment variables
load_dotenv()

# Get environment variables
ollama_model = os.getenv("OLLAMA_MODEL")

# Enable LiteLLM debugging
os.environ["LITELLM_LOG"] = "DEBUG"

# Register Ollama model with correct format
model_name = f"ollama/{ollama_model}"
litellm.register_model({
    model_name: {
        "api_base": "http://host.docker.internal:11434",
        "api_key": "not-needed",
        "api_type": "ollama"
    }
})

Saeed_Kasmani · February 4, 2025, 2:54am

I am reading it in another python file:

import streamlit as st
from crews import ResumeCrew
from pdf_utils import extract_text_from_pdf
from tempfile import NamedTemporaryFile

from dotenv import load_dotenv
import os
import fitz  # PyMuPDF for reading PDFs
import litellm

os.environ["CREWAI_TELEMETRY"] = "false"
os.environ["CREWAI_ANALYTICS"] = "false"
litellm.set_verbose=True
# os.environ.pop("OPENAI_API_KEY", None)
os.environ['LITELLM_LOG'] = 'DEBUG'
load_dotenv('creds.env')

Guigui_DuSud · February 6, 2025, 9:07am

see the thread that looks the most advanced with CrewAI team in String Knowledge sources not working with Gemini

eacee · February 11, 2025, 1:13am

Saeed_Kasmani:

import streamlit as st
from crews import ResumeCrew
from pdf_utils import extract_text_from_pdf
from tempfile import NamedTemporaryFile

from dotenv import load_dotenv
import os
import fitz  # PyMuPDF for reading PDFs
import litellm

os.environ["CREWAI_TELEMETRY"] = "false"
os.environ["CREWAI_ANALYTICS"] = "false"
litellm.set_verbose=True
# os.environ.pop("OPENAI_API_KEY", None)
os.environ['LITELLM_LOG'] = 'DEBUG'
load_dotenv('creds.env')

Have you tried adding load_dotenv(‘creds.env’, override=True) to override the standard .env implementation? Openly, I’m not familiar with how you have that implemented…

Renato_Guilherme · February 16, 2025, 9:15pm

Hi @Saeed_Kasmani !

Give a try on:

using dotenv do load environment variables from .env file
ensure to set OPENAI_API_KEY=NA
no need to have api_key in LLM object in case you are using local

Let us know if that worked or any issue.

Saeed_Kasmani · February 22, 2025, 1:01pm

Hi @Renato_Guilherme,
Thanks for your reply. i think there is an issue with using crewai, ollama and pydantic. I requested for help. and shared my code!

Zarul · March 13, 2025, 5:19am

from crewai import Agent, Crew, Process, Task, LLM
from crewai.project import CrewBase, agent, crew, task
from crewai_tools import SerperDevTool, ScrapeWebsiteTool, PDFSearchTool
from crewai.knowledge.source.pdf_knowledge_source import PDFKnowledgeSource
from models import (
    JobRequirements,
    ResumeOptimization,
    CompanyResearch
)

from dotenv import load_dotenv

load_dotenv()


@CrewBase
class Optimizer():
	"""Optimizer crew"""

	agents_config = 'config/agents.yaml'
	tasks_config = 'config/tasks.yaml'

	def __init__(self) -> None:
		self.resume_pdf = PDFKnowledgeSource(file_paths="CV_Mohan.pdf")

	@agent
	def resume_analyzer(self) -> Agent:
		return Agent(
			config=self.agents_config['resume_analyzer'],
			verbose=True,
			knowledge_sources=[self.resume_pdf],
			embedder=dict(
				provider="ollama",
				config=dict(
					model="nomic-embed-text",
					base_url="http://localhost:11434"
				)
			)
		)

	@agent
	def job_analyzer(self) -> Agent:
		return Agent(
			config=self.agents_config['job_analyzer'],
			verbose=True,
			tools=[ScrapeWebsiteTool()],
		)
	
	@agent
	def company_researcher(self) -> Agent:
		return Agent(
			config=self.agents_config['company_researcher'],
			verbose=True,
			tools=[SerperDevTool()],
			knowledge_sources=[self.resume_pdf],
			embedder=dict(
				provider="ollama",
				config=dict(
					model="nomic-embed-text",
					base_url="http://localhost:11434"
				)
			)
		)

	@agent
	def resume_writer(self) -> Agent:
		return Agent(
			config=self.agents_config['resume_writer'],
			verbose=True,
		)
	
	@agent
	def report_generator(self) -> Agent:
		return Agent(
			config=self.agents_config['report_generator'],
			verbose=True,
		)

	@task
	def analyze_job_task(self) -> Task:
		return Task(
			config=self.tasks_config['analyze_job_task'],
			output_file='output/job_analysis.json',
			output_pydantic=JobRequirements
		)

	@task
	def optimize_resume_task(self) -> Task:
		return Task(
			config=self.tasks_config['optimize_resume_task'],
			output_file='output/resume_optimization.json',
			output_pydantic=ResumeOptimization
		)
	
	@task
	def research_company_task(self) -> Task:
		return Task(
			config=self.tasks_config['research_company_task'],
			output_file='output/company_research.json',
			output_pydantic=CompanyResearch
		)

	@task
	def generate_resume_task(self) -> Task:
		return Task(
			config=self.tasks_config['generate_resume_task'],
			output_file='output/optimized_resume.md',
		)
	
	@task
	def generate_report_task(self) -> Task:
		return Task(
			config=self.tasks_config['generate_report_task'],
			output_file='output/final_report.md'
		)

	@crew
	def crew(self) -> Crew:
		return Crew(
			agents=self.agents,
			tasks=self.tasks,
			process=Process.sequential,
			verbose=True,
			knowledge_sources=[self.resume_pdf],
		)

I manage to handle the issue with OpenAI API key issue… I’m using llama3.2 btw… unfortunately I encountered an error with ScrapeWebsiteTool… if anyone manage to run this project locally, do share your valuable insight… thanks!

Topic		Replies	Views
Llm connection to local server CrewAI Community Support	2	81	March 17, 2025
Unable to Initiate Crew using Local Ollama General	5	575	October 29, 2024
CrewAI is not allowing the use of a knowledge source when using an Ollama-based LLM. It defaults to requiring an OpenAI API key, even though Ollama is correctly configure General openai , agent , crewai	6	483	February 6, 2025
Crewai and LM studio authentication issue General openai , crewai	4	270	December 17, 2024
Connecting Ollama with crewai Crews crewai	10	4445	February 15, 2025

I want to use the local LLM via Ollama, but I'm facing some issues. It keeps asking me for the OpenAI API key

Related topics