Project Settings

Workspace, agent instructions, usage and quotas, retrieval features, danger zone.

Project Settings is the per-project admin surface for Chat with Your Documents. It lets you rename the project, write the agent instructions every new chat starts with, see how much storage and document count the project is using, see which retrieval features your workspace has on, and (for owners and admins) delete the entire project.

Reaching this page

The Project Settings entry sits at the bottom of the Chat with Your Documents sidebar, below Chats. The topbar breadcrumb reads Chat with Your Documents / {project name} / Project settings.

Page header:

Element	What you see
Title	Project settings
Subtitle	Rename the project, check usage, and see which retrieval features are enabled.

The body is a single column with four sections: Workspace, Agent instructions, Usage and quotas, Retrieval. The fifth section, Danger zone, only appears for owners and admins.

Workspace

A card with a server icon and the heading Workspace.

Read mode (default)

Field	What you see
Project name label	Small uppercase caption.
Project name value	Bold, single-line, truncates if very long. End-edge Edit button with a pencil icon.
Description label	Small uppercase caption.
Description value	Multi-line text. When empty, shows the italic hint No description yet.
Project ID	A small caption above the project’s internal identifier rendered in a monospaced, left-to-right font.
Created	The creation date in the locale format (English: e.g. Apr 12, 2026; Arabic: e.g. ١٢ أبريل ٢٠٢٦).

Edit mode

Clicking Edit swaps the name and description into editable fields and reveals Cancel and Save buttons.

Element	Notes
Name input	Plain text, capped at 100 characters. Saves on Enter. Cancels on Escape.
Description text area	3-row text area, capped at 500 characters. Placeholder: What lives inside this project? Cancels on Escape.
Cancel button	Closes edit mode and discards changes. Disabled while saving.
Save button	Persists the new values. While saving, label is Saving with a spinner.

If neither the name nor the description changed, Save quietly closes edit mode without firing a request.

Toast on success: Project updated. Toast on failure: Couldn’t update the project. Try again.

While the project is still loading, the card is replaced by an animated placeholder.

Agent instructions

A card with the heading Agent instructions and the description Customise how the agent behaves for all new chats in this project. Existing chats keep whatever instructions were active when they were created.

Element	Notes
Label	Agent instructions (optional)
Helper text	Applied to every new chat in this project. Security rules always take precedence. Max 4000 characters.
Text area	Up to 4,000 characters. Placeholder: e.g. You are an HR policy assistant. Always cite the source document. Keep answers concise.
Save button	End-edge button. Disabled until the value differs from what is saved. Toast on success: Agent instructions saved. Toast on failure: Couldn’t save instructions. Try again.

Existing chats keep whatever instructions were active when they were created. Updating here only affects new sessions.

Usage and quotas

A card with the heading Usage and quotas and the subtitle Spend resets on the first of each month. Storage and documents are cumulative.

The card renders a small grid of meters.

Meter	What it shows
Storage	Total bytes across every knowledge base and document. Bar fills proportionally to the workspace’s storage cap.
Documents	Document count across the project. Bar fills proportionally to the document cap.
Monthly LLM spend	Spend in your workspace’s currency for the current month. When the meter is not yet wired, the value reads Coming soon and the bar is hidden.

When a quota is unset for your tenant the meter reads Unlimited and the bar is hidden.

Toast or inline error on failure: Couldn’t load usage.

While the data is loading, the card is replaced by an animated placeholder.

Retrieval

A card with the heading Retrieval and the subtitle Server-side capabilities configured by your workspace. Read-only here.

The body has two parts: a feature list and a default-models strip.

Features

Each row is a small card with a status pill (On or Off) on the end edge.

Feature	Description
HyDE query rewriting	Generates a hypothetical answer before the vector search. Higher recall on obscure phrasings, costs one extra small-model call per turn.
Query decomposition	Splits multi-hop questions into up to three sub-queries that run in parallel.
RAPTOR hierarchical retrieval	Builds a summary tree over chunks during ingestion so broad questions can return high-level answers.
OCR for scanned PDFs	Arabic-optimised Tesseract with page-rotation detection. Fires automatically on pages with very little extractable text.

Default models

A small strip with three labelled chips:

Label	Notes
Embedding	The embedding model the workspace uses for retrieval.
Chat model	The model the agent uses to write answers.
Router	The small model that decides between RAG, SQL, or both.

Toast or inline error on failure: Couldn’t load retrieval configuration. While loading, the card is an animated placeholder.

Retrieval features are read-only here. They are flipped at the workspace admin tier; ask your workspace owner if you need a feature on or off.

Danger zone

This section is hidden unless your role can create projects (owners and admins).

Element	What you see
Heading	Danger zone, in the destructive (red) color.
Description	Deleting a project removes its knowledge bases, SQL sources, and conversations. This cannot be undone.
Delete project button	Outline-style with a trash icon, red text and red border. Opens the delete dialog.

Delete dialog

Element	What it says
Title	Delete this project?
Body	This will permanently remove the project and everything in it: knowledge bases, documents, chunks, vectors, SQL sources, semantic terms, and every conversation. This cannot be undone.
Targeted name	The current project’s name shown in bold below the body.
Confirmation hint	Type the project name to confirm.
Confirmation input	Plain text input. The dialog blocks confirm until your input matches the project name byte-for-byte.
Cancel button	Closes the dialog. Disabled while a delete is in flight.
Delete project button	Destructive (red, white text) per the global rule. Disabled until the input matches. While deleting, label becomes Deleting with a spinner.

On success, you are routed back to the project picker landing surface. Toast: Project deleted. Toast on failure: Couldn’t delete the project. Try again.

Permissions

Action	Owner	Admin	Analyst	Viewer
Open this page	Yes	Yes	Yes	Yes
Edit project name and description	Yes	Yes	No	No
Edit agent instructions	Yes	Yes	No	No
See usage and retrieval	Yes	Yes	Yes	Yes
Delete project	Yes	Yes	No	No

Analysts and viewers see Project settings with the Workspace and Agent instructions cards in a read-only state. The Edit and Save controls remain visible but rejected actions surface a toast carrying the backend’s verbatim message. The Danger zone is hidden entirely so the wrong affordance does not appear.

Common questions

I clicked Edit and changed the name, but Save did nothing.

The button quietly closes edit mode if neither the name nor the description differs from what is currently saved. If you want to revert and stop editing, click Cancel explicitly. Saving with the same value is a no-op by design.

The Agent instructions text area says 'Max 4000 characters.' Can I exceed that?

No. The input is capped at 4,000 characters and additional keystrokes are dropped. If you need to feed the agent more context than that, attach a document to the project’s knowledge base instead of stuffing it into the prompt.

Updating the agent instructions did not change my existing chat.

That is intentional. Each chat session captures the instructions in effect when it was created. Updating the project’s instructions only affects new sessions you start afterward. If you want the new instructions in the old chat, start a new session.

The Monthly LLM spend meter says 'Coming soon.'

The spend meter is being rolled out and is not yet active in every workspace. Storage and document quotas continue to track normally; the spend meter will surface a real value once the metering pipeline is on for your tenant.

Why are the retrieval features shown but not editable?

Retrieval features (HyDE, query decomposition, RAPTOR, OCR) are workspace-level decisions that apply across every project. They are flipped at the workspace admin tier so the project owner cannot accidentally turn off a guarantee that other projects rely on. Ask your workspace owner to flip a feature for everyone.

Delete project is greyed out.

Two reasons can disable it: your role does not include project delete (analysts and viewers do not see the Danger zone at all), or you have not typed the project name correctly into the confirmation input. The match is byte-for-byte and case-sensitive; copy the project name from the Workspace card if you are unsure.

I deleted a project and now I cannot find anything that was in it.

Project deletion is a hard delete: knowledge bases, documents, chunks, vectors, SQL connections, semantic terms, and every chat session are removed at once. There is no trash for projects. Restore is not possible from the UI.

I see Couldn't update the project. Try again. but I have permission.

The toast carries the backend’s verbatim message; transient network errors are the most common cause. If the toast keeps repeating, refresh the page (your role may have changed mid-session) and try again.

Where to go next

Knowledge Bases

Manage the documents in this project.

SQL Sources

Review connected databases.

Chats

See every conversation in this project.

Share Links

Public chatbots and conversation snapshots.