CLI

raglight chat, raglight agentic-chat, raglight serve — interactive RAG from your terminal.

CLI

RAGLight ships with a full command-line interface. Three commands cover the main use cases:

Command	Description
raglight chat	Interactive RAG chat session in your terminal
raglight agentic-chat	Same as chat, but with the Agentic RAG pipeline
raglight serve	Deploy as a REST API (optionally with a Streamlit UI)

---

raglight chat

Starts an interactive terminal chat session backed by a RAG pipeline. On first launch, a setup wizard guides you through choosing your vector store, embeddings, and LLM. Subsequent runs can skip the wizard entirely via environment variables.

raglight chat

The wizard walks you through:

1. Vector database — Chroma or Qdrant, local path or remote host
2. Embeddings — provider and model
3. LLM — provider, model, and API base URL
4. Knowledge source — local folder or GitHub repository
5. Indexing — option to reuse an existing index
6. Chat loop — type your questions, get streamed responses

Responses are rendered as markdown in the terminal with streaming output.

Skip the wizard with env vars

Create a .env file (or export env vars) and the wizard is bypassed entirely:

.env

RAGLIGHT_LLM_PROVIDER=Ollama
RAGLIGHT_LLM_MODEL=llama3.1:8b
RAGLIGHT_EMBEDDINGS_PROVIDER=HuggingFace
RAGLIGHT_EMBEDDINGS_MODEL=all-MiniLM-L6-v2
RAGLIGHT_DB=Chroma
RAGLIGHT_PERSIST_DIR=./myDb
RAGLIGHT_COLLECTION=default
RAGLIGHT_DATA_PATH=./docs

Then just run:

raglight chat

RAGLight prints the active configuration and goes straight to the chat loop.

Commands in the chat loop

Input	Action
Any text	Send a question to the RAG pipeline
bye / exit / quit	End the session

---

raglight agentic-chat

Same as raglight chat, but uses the Agentic RAG pipeline — the LLM can call tools, reason in multiple steps, and go beyond simple retrieval.

raglight agentic-chat

The setup wizard is identical. The difference is in the pipeline: the agent decides when to retrieve, can combine multiple retrievals, and produces richer answers for complex questions.

Agentic mode requires an LLM that supports tool calling (e.g. llama3.1, gpt-4o, mistral-large).

See the Agentic RAG page for a full explanation of the pipeline.

---

raglight serve

Starts a FastAPI REST API exposing your RAG pipeline over HTTP. Entirely configured by environment variables — no Python code required.

raglight serve

Add --ui to also launch the Streamlit chat interface:

raglight serve --ui

See the REST API page for the full reference — endpoints, configuration variables, Docker Compose setup, and more.

CLI options

Option	Default	Description
--host	0.0.0.0	Host to bind
--port	8000	Port to listen on
--reload	false	Enable auto-reload (development)
--workers	1	Number of Uvicorn worker processes
--ui	false	Launch the Streamlit chat UI alongside the API
--ui-port	8501	Port for the Streamlit UI

---

Common environment variables

All three commands read the same RAGLIGHT_* environment variables:

Variable	Default	Description
RAGLIGHT_LLM_PROVIDER	Ollama	LLM provider
RAGLIGHT_LLM_MODEL	llama3	LLM model name
RAGLIGHT_LLM_API_BASE	http://localhost:11434	LLM API base URL
RAGLIGHT_EMBEDDINGS_PROVIDER	HuggingFace	Embeddings provider
RAGLIGHT_EMBEDDINGS_MODEL	all-MiniLM-L6-v2	Embeddings model
RAGLIGHT_DB	Chroma	Vector store backend (Chroma or Qdrant)
RAGLIGHT_PERSIST_DIR	./raglight_db	Local persistence directory
RAGLIGHT_COLLECTION	default	Collection name
RAGLIGHT_K	5	Number of documents retrieved per query
RAGLIGHT_DATA_PATH	—	Path to documents (skips wizard prompt)

---

Summary

• raglight chat — terminal RAG chat with streaming markdown output
• raglight agentic-chat — same but with tool-calling agent mode
• raglight serve — REST API; add --ui for the web chat interface
• All three share the same RAGLIGHT_* env vars — one .env file for everything