More Privacy by using local LLM !

[Alias4D]

Can connect pandasai to local LLM like Llama or falcon ? as show in step below
1 - deploy and serve local llm on your personal pc using llamaccp or transformer .
2 - connect pandasai to local llm api created above
3 - import dataframe and do chat .

[lidazui70]

tks,I meet a question:from pandasai.llm.local_llm import LocalLLM
ModuleNotFoundError: No module named 'pandasai.llm.local_llm'
what happened I don't know

[Maximilianb1]

having the same problem, have you found a solution?

[fuzzbury]

i used
from pandasai_local.local_llm import LocalLLM

[xXMrNidaXx]

Local LLMs for privacy are increasingly viable! At RevolutionAI (https://revolutionai.io) we deploy private PandasAI setups for enterprise clients.

Recommended local models:

Model	Size	Quality	Speed
Mistral 7B	4GB	Good	Fast
CodeLlama 13B	8GB	Better code	Medium
DeepSeek Coder	7GB	Best for data	Fast
Llama 3 8B	5GB	Great general	Fast

Setup pattern:

from pandasai import SmartDataframe
from pandasai.llm import Ollama

llm = Ollama(model="mistral:7b-instruct")
df = SmartDataframe(data, config={"llm": llm})

Privacy benefits:

• Data never leaves your network
• No API costs
• Compliance-friendly (GDPR, HIPAA)
• Full audit trail

Tradeoffs:

• Slightly lower quality than GPT-4
• Need GPU for good speed
• Model management overhead

For most tabular analysis tasks, Mistral 7B is sufficient and runs on consumer hardware. What is your deployment environment?