Blazing fast local AI Agent for Python coding
Project description
Nemo Agent
Nemo Agent is an expert Python AI software developer.
Features
- Runs blazing fast locally on an RTX 4070 or greater
- Generates Python project structures automatically using
poetry
- Writes Python code based on task descriptions
- Executes development tasks using AI-generated commands
- Utilizes the
mistral-nemo
language model for intelligent code generation - Implements best practices in Python development automatically
- Writes and runs passing tests up to 80%+ test coverage using
pytest
- Automatically fixes and styles code using
pylint
- Utilizes
git
to commit and track changes
Coding Ability
- Can solve some leetcode hards on some runs in about 30 seconds on an RTX 4090
- Can apply time complexity constraints in task requirements
Demo Video
Local Install
Requirements
- Python 3.9 or higher
- git
- curl
- Ollama running
mistral-nemo
- Ubuntu (22.04 or 24.04) with a minimum of an RTX 4070
Requirements Installation
- Python, curl, git are pre-installed on Ubuntu
- Ollama install instructions:
curl -fsSL https://ollama.com/install.sh | sh
ollama pull mistral-nemo
nemo-agent
install:pip install nemo-agent
- You are ready to use
nemo-agent
Cloud Install
Requirements
- RunPod account setup with your SSH and billing information
RunPod Setup
- Make sure you have setup your SSH keys
- Select a
4090
pod - Select the
RunPod Pytorch 2.1.1
template - Edit the template:
- Set
Container Disk
to 60 GB - Set
Expose HTTP Ports
to8888, 11434
- Add
environment variables
withOLLAMA_HOST
key and0.0.0.0
value
- Set
- Deploy your pod
- After deploying then login via SSH
- Run on the pod:
curl -fsSL https://ollama.com/install.sh | sh && ollama serve
- Run on the pod:
ollama pull mistral-nemo
- Run on the pod:
pip install nemo-agent
- You are ready to use
nemo-agent
Usage
After installation, you can use Nemo Agent from the command line:
nemo-agent "Your task description here"
If you run nemo-agent without any arguments, it will prompt you to enter a task.
Examples
Create a simple calculator:
nemo-agent "Create a simple calculator"
Generate a fizzbuzz program:
nemo-agent "Create a fizzbuzz script"
Limitations
- LLM codegen issues mean that web APIs like Flask and FastAPI are not supported.
- Due to the LLM getting confused with paths and files - writing 1 code and 1 test file is only supported.
Notes
- Tested all local models that can run on an RTX 4090 on Ollama other than
mistral-nemo
and none can be prompted to complete the task. - Tested with Mistral Nemo hosted API - the local/API models are not the same and the API version is very buggy with prompting
Contributing
Contributions to Nemo Agent are welcome! Please feel free to submit a Pull Request.
License
This project is licensed under the MIT License - see the LICENSE file for details.
Disclaimer
Nemo Agent generates code using AI. While it strives for accuracy and best practices, the generated code should be reviewed and tested before being used in a production environment.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for nemo_agent-1.0.1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 697462a235ee4aca5ef215490e9b705d6de4e4e76bd0b896998c7074007a5fd0 |
|
MD5 | 2ae2877843501a99c6592ba93587e7b6 |
|
BLAKE2b-256 | e9c57e8c697823a35c2267e5622fa9d90d3f1c32797f4c7256432dd01d375a9c |