AI Chat: HERC in a full-page conversation¶

A dedicated, full-width conversation with HERC — history, comfortable typing, and every feature of the floating chat — without leaving the module.

AI Chat and the floating panel are two windows onto the same assistant: the same composer, the same commands, the same conversations, the same HERC backend. The full page just gives you room to work.

SurfacesFull-page · floating panel

BackendOne HERC · shared store

Command keys/ · # · @

Accessagent-run role

When to choose this¶

Settle in for a long session

A full-width composer and scroll-back history, without leaving the module.
Resume where you left off

The lander lists active and archived conversations — open a row to continue.
Drive HERC with commands

Type / for saved prompts and # to reference catalog, glossary, contracts, or flows.
All of HERC, with room to work

Every feature of the floating chat, in a page you can spread out in.

When not to reach for the full-page AI Chat

A quick question while working elsewhere → use the floating chat (bottom-right) — it stays on the page you are on.
Transcription, provider keys, or experiments → those live under the AI section routes, not the chat lander.
Anything secret → don't paste credentials or secrets into a prompt.

Full-page vs floating panel¶

Both surfaces run the exact same chat: the same composer, command surface, message actions, and conversation store, all talking to the same HERC backend. Pick the surface that fits the moment, not the capability — they are equal.

Full-page AI ChatFloating chat

Where: the /chat lander, then /chat/conversations/… for a single thread.
Best for: focused, long sessions — a comfortable composer and full scroll-back history.
Extra: browse, open, archive, and restore past conversations from the lander table.

Where: bottom-right on every page — it stays on the page you are on.
Best for: a quick question while working elsewhere.
Same: same HERC, same conversations, same command surface.

flowchart LR
    FLOAT["Floating chat<br/>bottom-right · any page"] --> WS(["Shared chat workspace<br/>composer · commands · message actions"])
    PAGE["Full-page AI Chat<br/>/chat lander · /chat/conversations/…"] --> WS
    WS --> STORE[("Conversation store<br/>history · active / archived")]
    WS --> HERC(["HERC backend<br/>routed agents · governed tools"])
    HERC <--> STORE
    HERC -.reply.-> WS
    classDef accent fill:#ff8400,stroke:#e67700,color:#ffffff,font-weight:600;
    class HERC accent

Figure 1 — Two surfaces, one assistant. The floating panel and the full-page AI Chat share the same workspace, the same conversation store, and the same HERC backend — so a conversation reads and behaves the same from either surface.

Why teams reach for this

A long session with HERC deserves more than a corner panel. AI Chat gives you a full-width composer and the whole thread on screen, while still being the same assistant — the same conversations you started in the floating chat are here, and vice versa.

What you see¶

Surface	Where	What you see
Lander	`/chat`	KPIs, a creation trend, and a table of conversations (active / archived) — open a row to continue.
Thread	`/chat/conversations/…`	Full composer, message actions (helpful / not helpful, edit-and-rerun where available), and the command surface.

The command surface¶

Type a trigger character in the composer to open a picker.

Key	Opens	Notes
`/`	Saved prompts	Insert a reusable prompt.
`#`	Catalog · glossary · contracts · flows	Structured mention tokens.
`@`	Reserved	For future agent shortcuts.

Mentions are read-only

Mentions in the composer are structured tokens, and matching items are read-only — you navigate to them via links. Nothing mutates from the picker.

A message's lifecycle¶

Once you send a message, the thread walks it through send, reply, and the actions you can take afterwards.

flowchart LR
    COMPOSE["Compose<br/>/ prompts · # entities · @ reserved"] --> SEND["Send message"]
    SEND --> INFLIGHT{"In-flight<br/>request"}
    INFLIGHT -->|"you press Stop"| STOP["Cancelled in app<br/>provider may still finish outside Datahub"]
    INFLIGHT -->|"reply arrives"| REPLY["Assistant reply"]
    REPLY --> ACT["Message actions<br/>helpful / not helpful · edit-and-rerun"]
    ACT -.->|"edit-and-rerun"| RERUN["Rewrites the thread<br/>from that user message forward"]
    RERUN --> SEND
    classDef accent fill:#ff8400,stroke:#e67700,color:#ffffff,font-weight:600;
    class RERUN accent

Figure 2 — What happens to a message in a thread. Stop cancels the request from the app (a provider may still finish buffering outside Datahub in edge cases); edit-and-rerun rewrites the thread from the edited message forward rather than keeping parallel branch versions in the UI.

Permissions¶

Action	Who can do it
Run chat	Your role must include HERC / agent run — the same gate as using the assistant anywhere else.
Thumbs + optional feedback text	Same gate as sending messages.
Aggregated feedback for analytics	Admin / AI editor — see AI platform for organization settings.

Limitations¶

Limit	Why	What to do
Replies need a configured model / provider.	Chat runs on whatever provider your tenant has set up.	An admin adds one under AI platform. If nothing is available, the UI explains the gap — it is not a silent failure.
Edit-and-rerun replaces the tail of the thread.	It rewrites forward from that user message rather than keeping parallel branch "versions" in the UI.	The earlier tail is not kept as a separate branch — copy anything you want to preserve first.
Stop cancels the in-flight request from the app.	Provider-side buffering may still finish outside Datahub in edge cases.	Treat Stop as "stop showing me this," not a guarantee the provider halted.

Audit & compliance¶

Question a CISO might ask	Where to look
"Does chat respect our tenant and roles?"	Yes — feedback and chat usage follow your tenant and roles, the same gates as the rest of the platform.
"Where do provider keys and licensing live?"	Under AI platform.
"Can we keep secrets out of prompts?"	It's policy, not a filter — do not put secrets in prompts.

Troubleshooting¶

Symptom	Likely cause	Fix
403 / can't see AI Chat	Your role doesn't allow agent / conversation access.	Ask an admin to grant the HERC / agent-run role.
Empty, or "no provider"	No provider key or model configured.	Admin: add provider keys and model config under AI platform.
Mention search slow	Large org / many entities.	Narrow with a `#type:` prefix or a shorter query.