Skip to main content

AI Services module

A contextual chatbot, AI-powered insights, mentorship suggestions, and digests — backed by your choice of Anthropic, OpenAI, or Google Gemini, with bring-your-own-model support.

Optional module Enable in Settings → Modules

AI Services is the multi-provider AI layer. It adds a floating chatbot that can answer questions about your org's data, plus indexing and retrieval for documents, members, forum posts, and compliance records. Every org can use the platform default provider, bring its own API key, or fall back to keyword search when no provider is available.

Who this module is for​

  • Members — asking the chatbot questions about events, dues, documents, and the directory
  • Officers — quick lookups during meetings and recruitment events
  • Org admins — choosing a provider, scoping what the AI can read, monitoring the embedding index
  • Platform admins — configuring the platform-wide default provider and re-indexing tenants when models change

What gets enabled​

When AI Services is turned on for an org:

  • The floating chatbot widget appears in the bottom-right of every page for signed-in users and streams answers in real time
  • Content embeddings are generated for the data sources the org opts in to, on a nightly schedule and on demand
  • Org admins gain AI Services settings — provider selection, model name, an encrypted API key field, content-scope toggles, and an on-demand re-index button
  • AI features in other modules — onboarding suggestions, mentorship matching, notification digests, admin reports — light up when their parent modules are also enabled

Bring-your-own-model​

Every tenant can run AI on a different provider. The platform supports three:

  • Anthropic Claude — chat only; Anthropic does not currently expose an embeddings API, so embedding-backed retrieval falls back to keyword and trigram search when Anthropic is the active provider
  • OpenAI — chat plus embeddings (1536-dimension vectors)
  • Google Gemini — chat plus embeddings (768-dimension vectors)

The provider that actually answers a question is resolved through a three-step priority chain: platform default first, then the org's own override if it has one, and a platform fallback if both are unavailable. Each level can be marked active or inactive independently.

Content scope​

Content-scope toggles decide what the chatbot is allowed to retrieve. Four are surfaced in the org admin panel today — members, forum posts, documents, and compliance. Six additional scopes (events, finances, learning, foundation, alumni, elections) exist in the data model but are not yet wired into the org settings UI; they inherit their on/off state from module enablement until those toggles are added.

Common workflows​

What's not in this module today​

  • Per-response source citations. The chatbot answers in prose; it does not yet attach inline links to the specific record it drew from.
  • Thumbs-up / thumbs-down feedback on individual responses. No per-message rating is captured.
  • Persisted conversation history. The chat session is in-memory for the browser session. Closing the chat clears the thread.

Where to go deeper​

Dependencies​

  • A configured provider key — platform default, org override, or both
  • Storage — for embedding artifacts (platform-managed)
  • The modules whose data the AI reads — toggling Documents off, for example, removes documents from answers
  • Documents — indexed for chatbot retrieval when both are enabled
  • Community — forum posts feed the chatbot if the matching content scope is on
  • Operations — the chatbot can answer dues, election, and compliance questions

Last verified against v0.62.1 (2026-05-11).