Sections
Getting started
CLI flags & config
SQL
DDLDMLSELECTFunctions & expressionsTransactions
Clients
HTTP APINative driversPostgreSQL / MySQL wire protocolBinary RPC
Features
RBAC (multi-user)Embeddings (semantic search)File importBackupsDemo showcase mode and .fdb management
Operations
DeploymentArchitectureFileDB on Windows: tray and autostart
Reference
FileDB vs PostgreSQL / MySQL
AI
AI Features: overviewAI Chat and AI AgentSmart Export and AI Import MappingEmbeddings and Hybrid SearchAI setup: providers and presets
Dashboards
Dashboards: overviewDatasets and chartsNo-Code query builderDashboards AI assistantWidgets, grid and filtersPublic links and exportLayout, text cards and KPI comparison

AI Features: overview

FileDB isn't just a SQL engine with a built-in HTTP API — on top of it sits a set of AI features that work with any chat-completions-compatible LLM provider (OpenAI and OpenAI-compatible proxies, plus Voyage/Gemini/Yandex separately for embeddings). Everything is a plain JSON HTTP endpoint — no SDK, no extra dependencies in the engine itself.

Feature Endpoint What it does
AI Chat POST /api/ai/chat Conversational SQL assistant: replies with text and suggests SQL blocks, never runs them itself
AI Chat (stream) POST /api/ai/chat/stream Same thing, streamed
AI Agent POST /api/ai/agent Multi-turn read-only agent with tool calling — explores schema and data on its own
Smart Export POST /api/ai/sql-suggest Natural-language description → a ready SELECT for export
AI import mapping POST /api/import/ai-suggest Suggests a column mapping between the uploaded file and the target table
Embeddings /api/embed/* Vectorizes table rows, semantic (meaning-based) search
Hybrid search POST /api/vector/search Semantic search combined with a SQL WHERE filter in one call

One provider for the conversational features

AI Chat, AI Agent and Smart Export share one global named preset for the /chat/completions connection — configured once in the web UI and reused by all three. Details on the "AI setup: providers and presets" page.

Embeddings — configured per database

Unlike chat, the embeddings configuration (provider, model, key) is stored separately for every .fdb file — vectors are tightly coupled to whatever produced them, so different databases on the same server can use different embedding providers.

Safe by default

None of the AI features can silently damage data:

  • The AI Agent is read-only — only list_tables, describe_table, and run_select exist as tools; there is no DML/DDL tool for it to call, and an iteration cap (default 6) keeps it from running away.
  • AI Chat and Smart Export never execute SQL themselves — the model proposes code, the operator explicitly clicks "Run" after reviewing/editing it.
  • AI import mapping quietly falls back to heuristic name-matching if the LLM is unavailable (no key, network failure, malformed reply) — import never breaks because AI is down.

More detail on the "AI Chat and AI Agent", "Smart Export and AI Import Mapping", "Embeddings and Hybrid Search", and "AI setup: providers and presets" pages.

Related pages
← Previous
FileDB vs PostgreSQL / MySQL
Next →
AI Chat and AI Agent