Overview

The Ask page (/dashboard/ask) lets you chat with your project memory using natural language. Ask a question, and the system searches your memory items, finds the most relevant sources, and streams an AI-synthesized answer with citations.

Getting started

1. Navigate to the Ask page in the dashboard
2. Select your project from the project switcher (if not already selected)
3. Type a question in the input bar or click one of the suggested prompts
4. Press Enter (or click the send button) to submit

If this is your first question, a new chat session is created automatically and titled with your question.

Chat interface

Message input

The input bar is at the bottom of the page. It supports:

• Single-line submit -- Press Enter to send
• Multi-line input -- Press Shift+Enter for a new line
• Stop streaming -- Click the red stop button while an answer is being generated
• Retry -- If an answer fails, click the Retry button on the error message

Suggested prompts

When starting a new conversation (empty message list), four suggested prompts are displayed:

• "How does authentication work?"
• "What are the coding conventions?"
• "What is the project architecture?"
• "Are there any security issues?"

Clicking a suggestion populates the input bar. You can edit it before sending or submit it as-is.

Message bubbles

Messages are displayed as a conversation thread:

• User messages -- Displayed on the right with an orange accent background
• Assistant messages -- Displayed on the left with a purple AI avatar, rendered as full markdown (headings, code blocks, tables, lists)
• Error messages -- Displayed with a red border and an error icon, with a Retry button

While the assistant is generating, you see a "Thinking..." indicator followed by the streaming text appearing in real time.

Markdown rendering

Assistant responses are rendered with full markdown support:

• Headings (##, ###)
• Bold, italic, and inline code
• Fenced code blocks with syntax highlighting
• Tables
• Bullet and numbered lists

Session panel

The left sidebar shows your chat sessions. Toggle it with the panel button in the top-left corner.

Session list

Sessions are displayed in reverse chronological order (most recent first). Each session card shows:

Creating a new session

Click the New Chat button at the top of the session panel. This clears the current conversation and prepares a fresh session (which is created on the server when you send your first message).

Searching sessions

Use the search input at the top of the panel to filter sessions by title. The search requires at least 2 characters and uses full-text search on the session title.

Filtering favorites

Click the star icon next to the search bar to toggle between showing all sessions and showing only favorited sessions.

Favorite a session

Click the star icon on a session card to toggle its favorite status. Favorites are per-user -- other team members have their own favorites.

Deleting a session

Open the session detail modal (click the three-dot menu on a session card) and click Delete. This permanently removes the session and all its messages. Storage tokens used by the session are refunded to the team's rollup.

Source cards

Every assistant message includes source citations displayed below the answer. Sources are grouped into two categories:

Used Sources

Sources that the AI explicitly used to construct the answer. These are the most relevant sources and are marked with a green Used badge.

Related Sources

Sources that matched the question by semantic similarity but were not directly cited in the answer. These provide additional context if you want to explore further.

Source card details

Each source card displays:

Source card actions

Click a source card to expand it and reveal:

• Copy path -- Copy the schema key to clipboard
• Copy content -- Copy the facts text to clipboard
• View in Memory -- Navigate to the memory item in the Brain tab

Token info line

Below each assistant message, a token info line shows:

• Model -- The AI model used (e.g., gemini-2.0-flash)
• Tokens -- Prompt tokens, completion tokens, and total tokens used
• Sources -- Number of sources found

This helps you understand the cost and context of each answer.

Message feedback

Each assistant message has feedback buttons:

• Thumbs up -- Mark the answer as helpful
• Thumbs down -- Mark the answer as unhelpful
• Copy -- Copy the full markdown answer to clipboard

Session detail modal

Access the session detail modal from the three-dot menu on a session card. It shows:

• Title -- Editable session title (max 256 characters)
• Statistics -- Message count, total AI tokens used, storage tokens
• Model -- The model used in the session
• Created at -- When the session was started
• Delete -- Permanently remove the session and all messages

Tips for better answers

• Be specific -- "How does JWT refresh token rotation work in the auth middleware?" produces better results than "how does auth work?"
• Mention components or files -- "What does the useScrollAnimation hook do?" helps the semantic search find the right sources
• Ask about topics in your memory -- Ask works best for architecture, conventions, decisions, and implementation details that have been captured in memory items
• Use follow-up questions -- Build on previous answers in the same session for more focused exploration
• Check the sources -- If an answer seems incomplete, expand the source cards to see what information was available

Next steps

• Ask (Concept) -- How Ask works architecturally
• Ask Your Codebase -- Practical guide to getting the most out of Ask
• Memory -- Build the memory that powers Ask