OnField Assistant (ofa) 🌵

Locally-hosted, RAG-augmented LLM assistant for HPC and scientific-computing workflows. Runs Gemma 4 (31B, Q8_0, at its full 262144-token context) on a single GPU via Ollama and ships specialized modes for OpenFOAM, AMReX, MARBLES, ReFrame, VASP, quantum computing, HPC support, and general coding.

• Repository: github.com/nileshsawant/onfield-assistant

• Ships with: a VS Code extension that puts all eight modes in Copilot Chat’s model picker, a --serve OpenAI-compatible HTTP endpoint for other editors / opencode, a stdlib-only ofa_client Python module, and a one-command install.sh for standing ofa up on any Linux HPC.

What lives where

• User & porting guide: Getting started — end-to-end usage, CLI flags, installation on a new HPC, and the VS Code extension walkthrough.

• VS Code / other clients: Use ofa from VS Code — the manual BYOK route over an ssh tunnel. Superseded by the extension for Remote-SSH users; still the reference for other editors.

• Architecture: High-level layout and the in-depth technical overview.

• RAG maintenance: Updating the RAG corpora — operator playbook for git-tracked, vendored, and copyright-restricted corpora, plus the per-user private-data layer (ofa --add-private).

• API reference: Python API — every module, class, and function in src/, hyperlinked, with [source] jumps to the exact line on GitHub.