Configuration

ItihRag AI Chatbot is configured from ItihRag → Settings, organized into eight tabs. All forms are nonce-protected and every field is sanitized on save.

General

Enable or disable the plugin globally and set core options. This is also where the active LLM and embedding provider status is summarized.

LLM Providers

Choose your chat-completion provider: OpenAI, OpenAI-compatible, Anthropic Claude, Cloudflare Workers AI, Groq, or Ollama. Each provider has its own API key, base URL and model fields. Use Fetch models to populate the model dropdown from the provider, and Test connection to verify your credentials before saving. See Providers for per-provider notes.

Embeddings

Configure the embedding provider separately from the LLM. Supported: OpenAI, OpenAI-compatible, Cloudflare Workers AI, and local Ollama. The embedding dimension is auto-detected with a probe (cached in a transient); an explicit dimension override is available for providers that don’t report one.

Vector Database

The screen auto-detects whether your MySQL server is MySQL ≥ 9 with native VECTOR support. Choose an engine:

  • Auto — Vectorize when configured, otherwise MySQL.
  • MySQL — native VECTOR(n) columns on MySQL 9, or JSON + PHP fallback on older MySQL / MariaDB.
  • Cloudflare Vectorize — hosted vector index; a Create / verify index button auto-detects the dimension from your embedding model.

See Vector stores for the full comparison.

Indexing

Control how content is split into chunks:

  • Chunk size — target size (in characters) of each chunk.
  • Overlap — how many characters of overlap between adjacent chunks.
  • Minimum chunk length — chunks shorter than this are dropped.
  • Post types — which WordPress post types are eligible for indexing.
  • Auto-index options — re-index on save_post, remove on wp_trash_post.

The chunker is sentence-aware and preserves formatting (no sanitize_text_field mangling).

Chat

Behavior and personality of the chatbot:

  • Bot name, welcome message, and a system prompt.
  • Citations toggle — append a Sources list built from retrieval metadata.
  • Reasoning / extended thinking toggle — surface the model’s reasoning in a collapsible panel.
  • Top-K — how many chunks to retrieve per query.
  • Similarity threshold — minimum cosine similarity for a chunk to be used.
  • Rate limiting — per-IP request window and max requests (via transients).

Appearance

Skin the frontend widget:

  • 4 preset themes — Light, Dark, Ocean, Sunset.
  • Per-color overrides — primary, header background, text, user/bot bubble colors, launcher color.
  • Custom logo and bot avatar upload.
  • Launcher position — corner placement of the floating bubble.

Theming happens entirely through --openrag-* custom properties on the #openrag-widget root — no leaked styles, no global selectors.

MCP

Connect to external MCP (Model Context Protocol) servers in client mode. Add, edit, delete, enable and disable servers; click Discover to cache each server’s tool list. Enabled servers’ tools are offered to the LLM via function calling. See the MCP guide.

Advanced

  • Remove data on uninstall — opt-in full cleanup. Off by default, so your data is preserved when the plugin is deleted.