Panel LexiChat

Use the Chat page to work with a running model, attach private documents, manage conversation settings and save reusable workspace items. The panel does not start an engine just because a model is selected.

Start with a document and system prompt

1. Open Chat at /rust/chat and select a running model.
2. Open Chat settings, choose System prompt, and enter the instructions for this conversation.
3. Use Attach document to upload a local document; uploading does not send the message.
4. Enter a question, then choose Send message. Watch the status; a queued or rejected request is not a completed answer.
5. For an automated setup, use the Start a chat with a document and instructions guided task. Supply the document's base64 bytes, title, filename, document type and system prompt; the task creates the private chat and shows its identifier. Then open it in Chat and send the first message.

Native navigation, streaming and no-script operation

The Chat page uses the same Panel header, navigation, language controls and appearance selection as the other native pages. The conversation list appears beside the message history; the composer, running-model picker and Chat settings are in the chat workspace. The selected model is not started automatically.

With scripts enabled, Send requests progressive output. New text appears within the selected conversation as it arrives; Stop requests cancellation and the saved message status identifies an incomplete or stopped response. The page does not treat a transport failure as a completed answer. Attachments, tools, sharing, search, folders and saved workspace choices use the authenticated chat endpoints of the same page.

With scripts disabled:

1. Open Chat and choose a running model from Models. If none is available, start one from Manage running engines and reload.
2. Enter text in Message your model. Optionally expand the visible Chat settings section and set System prompt, Temperature, Top-p, Top-k, Min-p, Maximum tokens, Repeat penalty, Seed, Stop sequences, Web research or the supported TokenGate keep budget.
3. Choose Send. The server validates the model and settings, processes the response and reloads the saved conversation. If a request is rejected, correct the reported setting or model choice before retrying.
4. Read the saved reply in the message history. Reload the page to check the latest saved status if a prior generation was interrupted. Only the incremental display and browser pickers require scripts; sending and reading a saved reply do not.

The no-script form does not upload files or provide real-time token updates. Use an enabled browser for local file attachment and richer interactive controls.

Manage workspaces and preferences

1. In Workspace library at /rust/chat/workspace, create or edit a model preset, reusable prompt, note or text skill. Workspace roles and Workspace users control explicit sharing.
2. In My chat preferences at /rust/chat/preferences, adjust your model, theme, language, memory and voice settings. Choose Save preferences to apply changes.
3. Share link publishes a saved conversation snapshot; use Revoke share when access is no longer needed. A private conversation is not shared without this action.

Preview and prune old conversations

1. An administrator opens Chat administration at /rust/chat/admin and sets Retention days. Zero disables deletion.
2. Open Chat retention at /rust/chat/retention. Select Dry-run retention, review scanned, candidates, removed and skipped counts.
3. If the report is correct, confirm I authorize permanent deletion of eligible expired saved chats and select Prune expired chats. A changed preview refuses execution and requires another dry run. Deletion cannot be undone.
4. The guided task Safely prune expired chats runs the same protected /api/chat/settings/retention route with a preview and confirmation.

Control reference

Related pages

See Panel Overview, Panel Models, Panel TokenGate and Reference Guided Tasks.