Nemo-cli

Nemo CLI is an autonomous terminal-based AI coding agent powered by NVIDIA open-source models and Nebius Token Factory for lightning-fast software development.

Comment

Inspiration

Modern developers spend too much context-switching between writing code, running builds, debugging terminal errors, and searching documentation. With the rise of high-performance open-source models and scalable inference platforms like the Nebius Token Factory, we wanted to build an autonomous developer tool that lives right where developers work: the terminal. Nemo CLI was born to bridge the gap between high-speed serverless inference and a native, privacy-first, terminal-centric AI agent loop.

What it does

Nemo CLI is an autonomous AI coding agent designed for fast software development. Running directly in your command line, it allows you to:

  • Write and Modify Code: Implement complex features, fix bugs, and execute seamless refactors through an interactive Ink-based React terminal UI.
  • Autonomous Execution: Plan steps, execute shell commands, read files, and iterate based on real-time feedback loops.
  • Bring Your Own Provider: Connect seamlessly to NVIDIA NIM, Ollama, Llama.cpp, or custom OpenAI-compatible endpoints with full support for running locally to ensure zero provider network traffic and complete privacy.

How we built it

  • Core Runtime & TUI: Built using TypeScript and Node.js (>= 18), utilizing an interactive terminal UI powered by Ink for smooth streaming, plan modes, and rich tool visualizations.
  • Inference & Acceleration: Integrated seamlessly with NVIDIA open-source models powered by the high-throughput serverless endpoints of Nebius Token Factory.
  • Security & Sandboxing: Implemented a deny-first execution policy, OS-level sandboxing, and tamper-evident hash-chained audit logs to ensure secure autonomous workflows.

Challenges we ran into

Balancing ultra-low latency token generation with rich, multi-step agentic planning inside a terminal UI was a major hurdle. Managing context window compaction (/compact), streaming tool calls smoothly without breaking terminal rendering, and establishing a robust deny-first safety policy required careful architectural tuning.

Accomplishments that we're proud of

  • Delivering a fully functioning, npm-distributable CLI agent (@shrijanagain/nemo-cli) complete with rich slash commands (/newchat, /sessions,/provider,/model`), custom subagent personas, and a polished Ink-based TUI.
  • Successfully integrating high-throughput cloud acceleration via Nebius Token Factory & build.nvidia.com while keeping local fallback options (Ollama / Llama.cpp) completely frictionless.

What we learned

Building a developer-first tool taught us how critical time-to-first-token and predictable output streaming are for command-line interfaces. Deep-diving into prompt-agent loops and execution policies also significantly sharpened our understanding of secure, local-first AI system design.

What's next for Nemo CLI

  • Expanding native IDE extensions (VS Code integration).
  • Enhancing multi-agent swarm task distribution and background workers.
  • Optimizing context caching for even faster multi-file repository refactoring.

Built With

Share this project:

Updates

Submission history