0rbis-7B (TerraBox)

Offline knowledge box: a local-first web UI designed to navigate massive datasets (Wikipedia dumps, maps tiles, documents, and local AI models) without internet.

Goals

• Fully offline usage, resilient to network outages.
• Designed for large-scale data (indexed search, paginated results, streaming viewers).
• Simple, fast UI in plain HTML/CSS/JS.
• Clear separation between UI shell and future local APIs.
• Multilingual translation for non-English content via local models.

Current State

• UI shell with sections for search, tools, and placeholders for data modules.
• Frontend calls local endpoints (offline backend):
  • POST /chat for local AI Q&A
  • POST /translate for local translation (future)

Suggested Local API (future)

• GET /status -> storage, indexes, dataset counts
• GET /search?q=...&page=...&filters=... -> paginated results
• GET /doc/:id -> document metadata and content
• GET /map/tiles/{z}/{x}/{y} -> offline tiles
• GET /health -> health check
• GET /models -> local model list
• POST /chat -> local chat
• POST /completion -> local completion
• GET /status -> local system stats
• GET /datasets -> list registered datasets
• GET /indexes -> list registered indexes
• POST /translate -> local translation (stub)
• POST /datasets/import -> register local datasets (stub)
• POST /indexes/create -> register index (stub)
• POST /vision/analyze -> analyze images (stub)
• POST /rag/query -> RAG query (stub)

Backend (Offline-Only)

Constraints:

• No Internet dependency
• No external APIs
• Single machine (LAN access optional)

Skeleton structure:

• backend/api/ FastAPI routes
• backend/inference/ local inference stub (llama.cpp later)
• backend/database/ SQLite storage
• backend/config/settings.yaml model config

Run locally:

• run.sh (Linux/macOS)
• run.bat (Windows)

Docs:

• GET / -> API OK
• GET /index -> local HTML landing page
• GET /docs -> Swagger UI
• GET /redoc -> ReDoc

Local Folders

• models/ place AI models here (not committed)
• data/ SQLite and dataset/index storage (local)

Model Manager

The backend now routes all inference through a model manager: API -> ModelManager -> Model -> Engine -> Model files.

Dependencies

• Base API: backend/requirements.txt
• AI stack (install later): backend/requirements-ai.txt

Structure

• index.html main shell + navigation
• mistral.html local AI chat
• onnx-trad.html local translation
• signet.html local bookmarks
• src/css/style.css shared styling
• src/javascript/ UI scripts