Work/§01 / Index AI · 2024–2025

Navi

An open-source AI assistant for the command line, built for cybersecurity work and shipped in CSI Linux.

Illustration of a Navi terminal session listing installed chips, including memory and settings, and the active memory session
Illustration. A Navi session, rebuilt from the chip definitions in the repository.

Navi is a command-line AI assistant for security work. It ships with CSI Linux and is open source under the AGPL-3.0 license. Everything it knows is extended through chips: small Python modules that each add a command, a description, and their own aliases, so contributors can add tools without touching the core.

I rebuilt the system architecture from the ground up to be modular, faster, and easier to read, and built the module management that the chips system runs on. Then I added the parts that make it useful offline:

  • A local model. Navi installs and runs a model through Ollama on the machine itself, with installers for Windows and macOS.
  • Chat memory. Named sessions keep conversational context, stored locally. The memory chip lists, creates, removes, and switches them.
  • Retrieval. A RAG system reads PDFs so the model can answer from your own documents.
  • Settings. A settings chip to review and change the configuration, plus token limits for retrieval and memory.

Between June 2024 and early 2025, I made 156 commits to the project, second only to its founder.

Next project → World of Haiku A cyberpunk game that teaches real hacking skills inside a simulated Linux desktop.