Skip to content

AI tools

AI Chat (full-page)

Use AI Chat when you want a dedicated, full-width conversation with HERC: history, typing comfort, and the same features as the floating chat — without leaving the module.

When to choose this

Need Use
Focused, long session AI Chat (/chat → open a conversation)
Quick question while working elsewhere Floating chat (bottom-right) — stays on the page you’re on
Transcription, keys, or experiments AI section routes — not the chat lander

Permissions

  • Run chat: your role must include HERC / agent run (same as using the assistant elsewhere).
  • Thumbs and optional feedback text: same gate as sending messages; aggregated feedback for analytics is admin / AI editor (see AI platform for organization settings).

What you see

  • Lander (/chat): KPIs, trend, and a table of conversations (active / archived) — open a row to continue.
  • Thread (/chat/conversations/…): full composer, message actions (e.g. helpful / not helpful, edit-and-rerun where available), and command surface (type / for saved prompts, # for catalog/glossary/contracts/flows, @ reserved for future agent shortcuts).

Mentions in the composer use structured tokens; matching items are read-only (navigate via links, no mutations from the picker).

Limitations

  • Replies require a configured model / provider (see AI platform). If nothing is available, the UI explains the gap — not a silent failure.
  • Edit-and-rerun replaces the tail of the thread from that user message; it does not keep parallel “versions” of a branch in the UI.
  • Stop cancels the in-flight request from the app; provider-side buffering may still finish outside Datahub in edge cases.

Audit & compliance

  • Feedback and chat usage follow your tenant and roles; do not put secrets in prompts. Provider keys and licensing are covered in AI platform.

Troubleshooting

Symptom What to check
403 / can’t see AI Chat Role must allow agent / conversation access.
Empty or “no provider” Admin: provider keys and model config (AI platform).
Mention search slow Large org / many entities; try a narrower #type: or shorter query.