One-shot chat turn
Run a single question against your vector index:
ragrails chat "How do I authenticate?" \
--vector-db qdrant \
--collection docs \
--url http://localhost:6333 \
--llm-provider openai \
--llm-model gpt-4o-mini
Multi-turn with history file
Pass --history-file to persist history across turns. The file is created on the first turn and updated after each turn.
ragrails chat "How do I authenticate?" \
--history-file files/chat/history.json \
--vector-db qdrant --collection docs --url http://localhost:6333 \
--llm-provider openai --llm-model gpt-4o-mini
ragrails chat "What about the second step?" \
--history-file files/chat/history.json \
--rewrite-query
Interactive REPL
Run chat with no query argument to start a full interactive session: a REPL that keeps history, streams answers, and can call tools.
Slash commands
Type these in the REPL to control the session live:
| Command | What it does |
|---|
/model | List available models, or /model gpt-4o to switch mid-conversation |
/stream on / /stream off | Toggle token-by-token streaming |
/tools | List the tools the assistant can call |
/debug | Show the retrieval trace for the last answer (scores, rewritten query, chunks used) |
/debug is the fastest way to see why an answer was weak. It shows what was retrieved, the rerank scores, and the query that was actually searched.
| Tool | What it does |
|---|
web_fetch | Fetches a web page as readable text, for live lookups beyond your indexed content |
api_call | Makes a live HTTP request to perform an action the user asks for (e.g. trigger a transfer, fetch account details) |
api_call makes real requests with the values you provide. The REPL asks for confirmation before each tool call.
Options
| Option | Default | Description |
|---|
QUERY | none | Question (positional). Omit to start interactive REPL |
--llm-provider | openai | LLM provider |
--llm-model | gpt-4o-mini | LLM model name |
--max-tokens | 1024 | LLM max output tokens |
--embedder-provider | voyage | Query embedding provider |
--embedder-model | voyage-3 | Query embedding model |
--vector-db | qdrant | Vector DB provider |
--collection | None | Collection name |
--url | None | Vector DB URL |
--persona | "" | System persona for the chat turn |
--history-file | None | JSON file to load and save history |
--rewrite-query | off | Rewrite the query before retrieval |
--rewrite-session-context | "" | Context hint for query rewriting |
--rerank | off | Rerank retrieved chunks |
--reranker | voyage | Reranker provider |
--reranker-model | rerank-2-lite | Reranker model |
--rerank-top-k | 5 | Chunks to keep after reranking |
--disable-intent-routing | off | Always run retrieval, even for small-talk queries |
--disable-history-compaction | off | Return full history without summarisation |
