How to Build CrewAI Agents with Deepseek R1 100% Locally

---

---

AI Summary

Summary of Video Transcript

• The video demonstrates how to build AI agents locally using Deep Seek.
• Deep Seek’s R1 model is cost-effective compared to OpenAI’s GPT-3.01, with a cost of $219permillionoutputtokensversus$60.
• Deep Seek R1 is open-source, with technical reports and distilled models available.
• The distilled models include smaller versions and 32 billion and 70 billion parameter models comparable to OpenAI’s mini model.

Detailed Instructions

1. Download OL Lama:
   • Visit ama.com (exact URL not provided in the transcript).
   • Download OL Lama for your machine and install it.
2. Pull the Deep Seek Model:
   • On OL Lama, select the Deep Seek DR1 model.
   • Choose a parameter model (e.g., 1.5 billion parameter model).
   • Copy the model name.
   • Open terminal (Command Terminal on Windows or Terminal on Mac).
   • Use the command O Lama pull <model_name> to pull the model locally.
3. Run the Model Locally:
   • Use the command AMA run <model_name> to start a two-way chat with the model.
4. Create AI Agents with Crew:
   • Open Cursor and create a new project.
   • Use the command crew AI create crew and name the crew (e.g., Deep Seek).
   • Clear default settings to prepare for local use.
5. Configure Crew for Local Use:
   • In crew.py, import llm from crew AI.
   • Create an instance of the llm class with the model and base URL.
   • Prepend the model name with “O Lama” to specify the local provider.
   • Pass the llm instance to the agents (e.g., researcher agent and reporting analyst agent).
6. Understand Light LLM:
   • Visit light lm.a (exact URL not provided in the transcript) for documentation.
   • For other providers, prepend the model name with the provider’s name.
7. Run the Crew:
   • In main.py, import the crew with from crew import deep seek.
   • Save all files after making changes.
   • Navigate to the correct directory in the terminal.
   • Run the crew with python Source SL deep seek slain.main.py.

Notes

• The video does not provide exact URLs for some steps.
• The video includes an offer to join a community for further learning and a master class for no-code platforms.
• The video concludes with the successful local execution of Deep Seek R1 with AI agents.