AI agent for RAG pipelines, LLM evals, prompt optimization, fine-tuning, and ML training. Works with Claude Code, Cursor, and VS Code IDE.
Used by ML engineers & researchers worldwide · Runs on your GPU cloud
Bring Neo into your editor, connect it to Claude Code over MCP, or run it in the cloud with zero setup. Pick how you want to build.
Install the Neo extension and delegate ML engineering — RAG systems, LLM evals, fine-tuning, and pipelines — without leaving VS Code or Cursor.
Neo
Works in VS Code, Cursor & Open VSX
NEO helps you build, test, and improve AI systems from a single task prompt.
It can research approaches, write code, run experiments, debug failures, benchmark models, evaluate outputs, and produce reports.
Instead of managing scripts, logs, evals, and manual fixes yourself, NEO handles the engineering loop and keeps iterating until the task is done.
And More...
Ask Neo to fix AI model training pipeline
Add new AI features in your brownfield projects
Analyze data leakage in your training pipeline
Use Neo with Claude Code or Codex
Neo uses multi-step reasoning with its extensive knowledge base and GPU sandbox to perform iterative ML experimentation — running 100s of experiments and automatically selecting the best model.
Take control via our interactive chat interface. Guide Neo's exploration of models and approaches, providing context and expertise to accelerate tasks.
Unlock Neo's full potential with multi-step reasoning. Neo proactively explores multiple approaches, assesses outcomes, and evaluates risks to find the most effective solution.
34.2%
#1 Score on MLEBench in Aug 2025
75
Competitions entered
#1
vs RD-Agent & AIDE on MLE-bench
10×
Faster ML development
As of August, 2025
Install the extension in VS Code or Cursor and give NEO your first ML engineering task.
Use cases
NEO helps with the AI engineering work behind modern AI products: model evals, prompt tests, RAG pipelines, dataset prep, experiments, and reports. Share the goal and context, then review, steer, and use the final outputs.
State the outcome in natural language. Fine-tune a model, ship an agent, or build a benchmark with no boilerplate prompt engineering.
Point NEO at your repo, data, connectors, and constraints so the plan fits the hardware and conventions you already run.
NEO writes the code, runs long experiments, evaluates, and hands back versioned artifacts for your review.
Replay on real scenarios, ask for sweeps, harden failure modes, and promote the winning run to staging when you are ready.
Evaluate & Benchmark
An async LLM benchmarking platform that evaluates models from OpenAI, Anthropic, Google, and more across 150+ real-world tasks covering coding, reasoning, structured output, and long-context retrieval.
Evaluate & Benchmark
Closed-loop system: an optimizer LLM writes prompts and reads failure summaries, a target LLM runs batches against synthetic data, and a JSON ledger tracks every iteration until scores converge.
Build Agents
10 specialized agents coordinating over async message bus: +4.62% returns across 250 days of S&P 500 data.
Install for Cursor