Agentic RAG (Retrieval-Augmented Generation) combines a document store, FAISS index, and multiple LLM adapters (Ollama, OpenAI Assistants, Google Vertex AI) so you can ask natural language questions over your own knowledge base.

• Local Ollama server + REST/MCP APIs for hybrid deployments
• Ollama Cloud support - Use cloud-hosted models with automatic local fallback
• Role-based startup (monolith, server, client) so you can choose which components to run
• Automatic grounding, citations, reranking, and verification powered by FAISS + Litellm
• React UI with file manager, logs, metrics, and backend switching
• Command-line helper (cli_agent.py) and Python APIs for scripting

git clone https://github.com/yourorg/agentic-rag.git
cd agentic-rag
# start all services (Ollama, REST API, MCP server, UI)
start.sh
# stop everything
stop.sh

start.sh/start.py do the heavy lifting:

1. create/activate .venv
2. install dependencies (pip install -r requirements.txt)
3. install the correct torch wheel per-platform (Intel → 2.2.2, others → PyTorch CPU wheel)
4. respect --skip-ollama, --skip-ui, and role flags when launching services

• Runtime settings (config/settings.json) override .env; edit them via the Settings UI to update Ollama endpoint, model, or logging behavior.
• Ollama Cloud: Configure cloud mode, API keys, and endpoints via Settings UI. See docs/ollama-cloud-architecture.md for details.
• Server-side details (MCP, REST, CLI) live in docs/models-and-configuration.md, docs/openai_assistants.md, and docs/google_integration.md.
• Desktop packaging notes are in docs/electron-desktop.md; API references are under docs/api_reference.md and docs/api_model_selection.md.
• Torch/NumPy handling, service logs, and advanced diagnostics belong in the same docs folder—refer there before editing the startup scripts.

# run the full pytest suite
.venv/bin/pytest tests/

# run linting
.venv/bin/pylint src/servers/rest_server.py src/servers/control_daemon.py src/core/factory.py

See docs/ for more targeted guides (ops, testing, API calls, cloud integrations).