Quick Start
Install Neo and complete your first end-to-end AI or ML task inside VS Code in under 5 minutes.
Neo reads your project structure, creates an execution plan, and completes real AI and ML tasks directly inside VS Code. From agentic workflows to traditional machine learning pipelines, Neo stays with the task from start to finish.
Installation Steps
Follow these steps to install Neo and run your first task.
Step 1: Create an Account
Sign up at heyneo.so to authenticate the VS Code extension.
Step 2: Install the VS Code Extension
Neo integrates with VS Code to read your project structure, execute code locally, and work with your Git workflow.
Step 3: Run Your First Task
Provide Neo with an AI or ML task, such as building agentic systems, fine-tuning models, or implementing RAG pipelines. Neo will plan the work, execute it locally, and iterate based on results.
Choosing Your Platform
đź’ˇ Recommended: Use the VS Code extension for full access to local files, code editing, and Git integration.
Multi-step tasks that span files, experiments, or iterations are best handled inside the IDE.
When to Use Neo
Neo is designed for AI and ML tasks that require multiple steps or extend beyond isolated code snippets.
Use Neo when you need to:
- Build agentic AI systems that plan, execute, and adapt autonomously
- Implement RAG pipelines with retrieval, generation, and evaluation components
- Fine-tune language models or vision models on custom datasets
- Analyze an unfamiliar ML project or repository structure
- Execute and iterate on multi-step workflows involving data, code, and evaluation
- Generate artifacts that you can inspect, modify, and commit to version control
- Work within your development environment without switching contexts
Neo is suited for both modern AI applications and traditional ML tasks that require planning, execution, and iterative refinement.
Common AI and ML Task Examples
Neo supports a wide range of modern AI applications and traditional ML engineering tasks. Here are examples to help you get started:
LLM Fine-tuning
Fine-tune a language model on custom instruction data and evaluate performance using domain-specific benchmarks
Model customization and domain adaptation
Agentic AI Workflows
Design an autonomous agent that researches a topic, summarizes findings, and tracks progress
Autonomous task execution and workflow automation
Retrieval-Augmented Generation (RAG)
Build a RAG pipeline over internal documentation and evaluate retrieval quality
Knowledge retrieval and document search
Computer Vision
Fine-tune a pretrained model on CIFAR-10 and compare validation performance
Image classification and object detection
Regression Analysis
Analyze the Boston Housing dataset, build a baseline regression model, and report key metrics
Predicting continuous values
Time Series Forecasting
Load historical stock prices, train a forecasting model, and evaluate the next 30-day forecast
Temporal predictions and trend analysis
Classification Models
Build and evaluate a sentiment classifier using the IMDB reviews dataset
Categorical outcomes and pattern recognition
Writing Effective Task Instructions
Neo responds best to task-oriented instructions that specify inputs, expected outputs, and success criteria, especially for multi-step workflows.
Example Instructions:
Build a RAG system using the company knowledge base, implement semantic search with embeddings, and evaluate retrieval accuracy.Fine-tune GPT-2 on customer support conversations and compare perplexity scores across different checkpoint epochs.Analyze the customer churn dataset, identify data quality issues, build a classification model with at least 80% accuracy, and generate a performance report.Review the model training code in src/models/train.py, identify bottlenecks, and refactor into modular functions with benchmarks.Best Practices
Be Specific About Inputs and Outputs
Include dataset names, file paths, model types, and expected artifacts. Example: “Use customer_data.csv to build a classification model and save plots as PNG files in the results folder.”
Define Success Criteria
Specify performance thresholds, metrics, or quality requirements. Example: “Generate a confusion matrix and ROC curve with AUC above 0.85.”
Break Down Complex Workflows
For large projects, use sequential steps. Example: “First, validate data quality. Then, engineer features. Finally, train and evaluate a baseline model.”
Reference Existing Code and Files
Point Neo to specific locations in your codebase. Example: “Refactor the training pipeline in src/train.py to support multiple model architectures.”