Using Eric
This page covers the user-facing Eric workspace at /eric.
Starting a conversation
Section titled “Starting a conversation”You can open Eric from:
- The message icon in the app header
- The Eric item in the main navigation
- A direct link to
/eric
If your company has exhausted its usable AI Credits, Eric shows a blocked state until an admin adds more capacity. See Assistant Settings for credit details.
Conversation list and session behavior
Section titled “Conversation list and session behavior”Eric stores work in conversations.
What the session list does
Section titled “What the session list does”- Shows saved conversations in reverse-updated order
- Lets you start a New conversation
- Lets you rename a conversation
- Automatically generates a title after the first exchange if you do not name the session yourself
- Prevents deleting a conversation while it still has an active run
- Shows a running indicator when Eric is still working in another session
Back behavior
Section titled “Back behavior”The Back button returns you to the last non-Eric page you came from. If you opened Eric directly, Back falls back to Dashboards.
Sending messages
Section titled “Sending messages”- Press Enter to send
- Press Shift+Enter to insert a new line
- Click the paper-plane button to send
If Eric is already responding and you send another prompt in the same conversation, the next prompt waits behind the active run instead of interrupting it.
While Eric is working
Section titled “While Eric is working”During a run, the send button changes into a Stop control.
You can:
- Watch streaming status updates and tool activity
- Cancel the active run with Stop
- Re-attach to a still-running response if the stream was interrupted and Eric offers a Re-attach button
Reading Eric’s responses
Section titled “Reading Eric’s responses”Eric responses can include several UI elements:
| UI element | What it means |
|---|---|
| Tool call cards | Expandable cards for each tool Eric ran, showing arguments, results, and status |
| Final answer cards | Structured answer blocks, including text and tabular outputs |
| Widget draft artifact cards | Quick links to open a drafted widget in the preview canvas |
| Knowledge Proposal notices | References to proposed Company Knowledge changes that approvers can review |
| Copy button | Copies the assistant message content |
| Helpful / Not helpful | Sends feedback tied to that assistant run |
Tool call cards
Section titled “Tool call cards”Each tool Eric runs appears as a card in the response. You can expand any card to see:
- The tool name and status — pending, completed, or error
- Arguments Eric passed to the tool
- Results, including dataset matches, SQL queries, row counts, and runtime warnings
- How long the tool took
Common tools you will see:
- Finding useful datasets — searches your accessible datasets and returns top matches with confidence scores
- Working with datasets — runs a SQL query against your workspace data and shows returned rows, columns, and datasets used
- Reading and updating Knowledge Files — inspects the Knowledge Filesystem and updates writable Markdown files
- Creating widget draft — builds a widget draft and returns layer counts, publish readiness, and a link to the canvas
If a tool fails, the card shows the error and any repair hint. You can copy the error or continue the conversation to have Eric retry.
Widget draft artifact cards
Section titled “Widget draft artifact cards”When Eric creates or updates a widget draft, a card appears in the chat thread. Click View in canvas to open the draft preview and publishing options.
Knowledge Proposals
Section titled “Knowledge Proposals”Eric can write proposed company-wide knowledge updates into the Knowledge Filesystem. Users with Company Knowledge Update Permission review those proposals in the Company Knowledge workspace, compare the proposed Markdown against the current file, and approve or reject the change.
Eric can update writable personal, scratch, and permitted Dataset knowledge directly while working. Public company-wide writes require the proposal workflow.
Knowledge in chat and workspace
Section titled “Knowledge in chat and workspace”Eric starts each run with a compact Knowledge Filesystem bootstrap. He can read visible Knowledge Files with exec_command and update writable Knowledge Files with apply_patch before responding.
You can inspect and maintain knowledge in the Company Knowledge workspace:
| Area | Who can edit it | How Eric changes it in chat |
|---|---|---|
| Personal Knowledge Directory | The signed-in user | Eric may update it directly when useful |
| Dataset Knowledge | Users with the required Dataset access and contribution permission | Eric may update it directly when permissions allow |
| Company Knowledge | Users with Company Knowledge Update Permission | Eric creates a proposal for approver review |
The workspace’s Map and Files views show these areas together according to your permissions. Files also exposes visible generated Markdown and live YAML Definition Files. Definition Files are read-only in Company Knowledge; use the provided action to open the Dashboard or Query Definition Workspace that owns one.
Credit limits
Section titled “Credit limits”Eric uses the Company’s Weekly AI Credit Allowance first. If that weekly allowance is exhausted and the Company has usable One-Off AI Credits, Main Eric Chat automatically uses those overflow credits without requiring additional user action. Admins can see this Overflow Credit Use in AI Assistant settings.
If the Company has no usable Weekly AI Credit Allowance or One-Off AI Credits, Eric shows an exhausted-state message and will not start new Main Eric Chat runs until more usable AI Credits are available. Non-admin Users are told to ask an admin for more AI Credits. Admins can buy One-Off AI Credits or manage an AI Add-On from Settings → AI Assistant → Billing & Credits or Settings → Subscription.
Weekly AI Credit Allowance resets weekly. One-Off AI Credits are separate purchased credits that roll over across weekly periods and expire one year after purchase.
Chat settings and credit state
Section titled “Chat settings and credit state”The Eric chat sidebar includes a settings entry point that shows the company’s current AI credit state. Anyone with Main Eric Chat access can open it.
What non-admins see
Section titled “What non-admins see”- A read-only Usage Summary with team activity, a leaderboard, and current weekly allowance usage
- AI Credit Availability, including remaining One-Off AI Credits and the next weekly reset time
- A compact low-credit indicator when total usable credits drop to 5% or less of the current weekly allowance
What admins see
Section titled “What admins see”Admins see all of the above, plus purchase and add-on management actions.
Low credits and exhausted states
Section titled “Low credits and exhausted states”- Low AI Credits Warning — appears in the chat sidebar when total usable credits are above zero but no more than 5% of the current Weekly AI Credit Allowance. You can still send messages; the warning is advisory and does not block an active Eric Run. You can collapse the expanded warning, but a compact indicator remains while credits are low.
- AI Credits Exhausted — blocks new message submissions. Non-admins see a prompt to ask an admin for more credits. Admins see purchase actions.
- Admitted run protection — if you send a message while credits are still available and the company runs out before Eric finishes, that run is allowed to complete. This overrun is recorded for admin visibility but does not create credit debt or reduce future allowance.
Trust and data boundaries
Section titled “Trust and data boundaries”Eric’s chat UI includes a trust strip with three important reminders:
- Resplendent does not train on your data
- Third-party AI providers are used with zero retention
- Eric can only use datasets you are already allowed to access
That last point matters most operationally: if you cannot reach a dataset in Resplendent, Eric cannot use it either.
Feedback and iteration
Section titled “Feedback and iteration”Use the feedback buttons on assistant messages to tell Eric whether a response was helpful.
This is best when:
- The answer was clearly correct or clearly wrong
- A tool failed
- The response was too slow
- The answer was unclear
Admins can go deeper than feedback by opening the Run Inspector and Trace Viewer.