Odysseus is the planned AI operations assistant for The Citadel: a local-first intelligence layer designed to search project documentation, explain system state, summarize operational notes, and support future monitoring workflows.
The goal of Odysseus is to give The Citadel a private assistant that understands the lab, the notes, the services, the project history, and the future roadmap. Instead of searching scattered files manually, Odysseus is intended to become a guide through the entire system.
Local-first. Private by default. Useful before flashy. The assistant should support the lab, not become the lab.
A retrieval-augmented generation workflow was built to search Citadel source-of-truth notes and return grounded answers from local project documentation.
ChromaDB and embeddings were tested as the memory layer. Documentation was ingested into a local vector database for semantic search.
Ollama models were tested locally, including smaller and larger models. OpenRouter was also tested for external model access during the prototype phase.
A command-line launcher was created so Odysseus could be queried directly from the Linux environment.
Web-based AI interfaces were tested, including Open WebUI and the Odysseus container stack.
Self-hosted AI expansion was paused because CPU-only performance was too slow for the desired experience.
The Citadel currently runs on a mini PC / Ubuntu VM environment with strong lab value, but no dedicated NVIDIA GPU.
Larger local models produced slow token speeds. The system worked, but not fast enough to justify making it the active center of the lab.
Odysseus remains a documented prototype. The AI layer will return when the hardware can support it properly.