Platform Chat

Chat with your ToraDB data — search, SQL analytics, charts, and reports.

Platform Chat

Platform Chat (/chat) is a data assistant grounded on your ToraDB tables. Ask natural-language questions; the assistant calls search and SQL tools, shows citations and charts, and can produce exportable reports.

Setup

Browser LLM (local development)

1. Open Settings in the platform.
2. Under LLM (Chat), choose a provider preset or enter:
   • API key — stored only in your browser (localStorage), never sent to ToraDB.
   • Base URL — e.g. https://api.openai.com/v1 or http://localhost:11434/v1 for Ollama.
   • Model — e.g. gpt-4o-mini.
3. Click Test LLM connection, then open Chat.

Local Ollama/LM Studio may require CORS on the LLM HTTP server when using browser mode.

Server proxy (production)

Configure the ingest/API server:

export TORADB_LLM_BASE_URL="https://api.openai.com/v1"
export TORADB_LLM_API_KEY="sk-…"
export TORADB_LLM_MODEL="gpt-4o-mini"

Restart toradb-ingest platform serve. In Settings, enable Use server proxy. API keys stay on the server; the browser only calls /api/chat/completions.

Vector search (optional)

Enable Embedder in Settings with an OpenAI-compatible /embeddings endpoint so Chat can use dense/hybrid retrieval on vector tables.

Example prompts

• “How many rows are in passages?”
• “Find documents about wireless power transfer.”
• “Count records by tag and show a chart.”
• “Write an executive report on the top themes in this table.”

Export

From the Chat header:

• MD — Markdown report of the thread (queries, tables, citations).
• HTML — Printable single-file report.

Tools the assistant uses

Tool	Purpose
list_tables	Table names and row counts
describe_table	Schema and samples
search	BM25 / hybrid / strategy-based retrieval
embed_query	Query embeddings (when embedder configured)
run_sql	Read-only analytics SQL
render_chart	Bar charts
generate_report	Structured report sections

SQL mutations (DROP, DELETE, CREATE, etc.) are blocked in Chat.