- Category
- AI Agents
- Rank
- No. 1280Tools index
Previous survey · No. 1287 ·
- Pricing
- Open Source
- Type
- AGENT
- Builder
- Hmbown
- GitHub
- 6 stars
- Latest release
- v0.1.26
- Date
About
Terminal-first control plane for NVIDIA Nemotron — agentic coding, RAG, doc-ops, and multi-model formations.
What it does
NeMoCode is a Python coding assistant that runs from an interactive terminal, full-screen interface, or one-shot command. It reads and searches projects, edits files, executes shell and Git operations, and delegates work to specialized agents. Models can come from hosted NVIDIA NIM, other OpenAI-compatible services, or locally served NVIDIA GPU backends.
Why it's ranked here
Its strongest case is unusually broad control over model placement and task routing. Read-only planning has an approval gate, while formations can assign planning, execution, and review to different models. The alpha classifier and community-project status temper that ambition, but the documented permission modes, diagnostics, endpoint management, and typed configuration show substantial operational scope.
What's good
The safety model is concrete: planning cannot write files, run shell commands, or commit, and execution starts only after approval. Local retrieval chunks project files, calculates cosine similarity without a separate vector database, and optionally reranks results. Custom agent profiles and markdown-defined skills let teams encode repeatable roles and workflows without changing the package.
Tradeoffs
Local model serving assumes NVIDIA GPU infrastructure plus vLLM, SGLang, TensorRT-LLM, or containers, and serving flags may vary between releases. Hosted use requires provider credentials. Automatic approval permits every tool action, so it demands care. The package declares alpha status, and the repository itself calls the project community-run rather than NVIDIA-endorsed.
How to use it well
It best suits terminal-oriented developers comparing hosted NIM with private, locally served models or splitting coding work across planner, executor, and reviewer roles. Start in read-only planning, inspect the proposed steps, then approve execution. Use project agent profiles and skills for recurring practices. It does not provide model hosting itself, replace required GPU serving software, or claim official NVIDIA support.
Technical notes+
pyproject.toml defines a Hatchling-built Python package under src/nemocode, requires Python 3.11+, and exposes nemo = "nemocode.cli.main:app". src/nemocode/cli/main.py registers the Typer command tree and launches the REPL when no subcommand is supplied. src/nemocode/config/__init__.py deep-merges bundled, user, project, nested-directory, markdown-agent, and environment configuration layers. src/nemocode/core/index.py implements FileIndex with overlapping 1,500-character chunks, batched embeddings, cosine scoring, and optional reranking. src/nemocode/tools/__init__.py generates OpenAI-format JSON schemas from type hints, coerces model-supplied arguments, and dispatches async tools through ToolRegistry. src/nemocode/skills/__init__.py discovers markdown skills and runs their prompts through the active agent.
Observed
- License
- MIT
- Primary language
- Python
- Runtime
- Python 3.11 or newer; classifiers list Python 3.11, 3.12, and 3.13
- Packaging
- PyPI package built with Hatchling; installable with pip, pipx, or editable source install
- Interfaces
- Typer CLI with interactive REPL, one-shot commands, piped input, and full-screen TUI
- Model interface
- Hosted NVIDIA NIM and other OpenAI-compatible APIs, plus local vLLM, SGLang, TensorRT-LLM, and NIM containers
- Extension surfaces
- Custom agent profiles, markdown-discovered skills, plugins, and MCP integration
- Project status
- Declared alpha software and a community project not affiliated with or endorsed by NVIDIA
Read from README.md, pyproject.toml, src/nemocode/__init__.py, src/nemocode/cli/main.py, src/nemocode/core/index.py, src/nemocode/cli/__init__.py, src/nemocode/core/__init__.py, src/nemocode/tools/__init__.py, src/nemocode/config/__init__.py, src/nemocode/skills/__init__.py, src/nemocode/providers/__init__.py, src/nemocode/workflows/__init__.py, src/nemocode/cli/commands/__init__.py, docs/INDEX.md, docs/UX_OVERHAUL_BRIEF.md.
What it can do
Generate code from natural language prompts
Natural language coding instructions → Generated code
Perform retrieval-augmented generation queries
Query text and knowledge base → Contextually relevant responses
Execute document operations and transformations
Documents and operation commands → Processed documents
Coordinate multiple AI models in formations
Multi-model configuration and tasks → Coordinated model responses
Control NVIDIA Nemotron models via terminal
Terminal commands and model parameters → Model execution results
Create agentic coding workflows
Coding task specifications → Automated code generation and execution
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.