In This Article: An orientation to the Smith tab in Compass — what it is, how the interface is organized, how to start your first conversation, and how to bring your Lakehouse files into context for grounded, document-aware responses.
What Is Smith?
Smith is Compass's built-in AI assistant, designed for financial analysis, strategic planning, document synthesis, and business intelligence work. It runs across multiple leading AI models simultaneously, so you're not locked into a single engine — you can choose the model best suited to the task, compare responses side by side, or let Smith select automatically.
Unlike general-purpose chatbots, Smith is built to work directly with your organization's files. Connect it to documents stored in your Lakehouse and Smith grounds its responses in your actual data, not generic training knowledge. Your files are never used to train AI models, and your conversations remain private to your organization.
Choose from multiple AI models — including Gemini, GPT, Claude, and DeepSeek — or select several for side-by-side comparison. Each model has different strengths in reasoning, speed, and context capacity. Newer GPT and Claude generations are available as advanced models once your project activates them, and multi-model is off by default.
Select files or folders from your Lakehouse to ground responses in your organization's documents. Smith can synthesize, analyze, and generate new content based on your proprietary data.
Configure Smith's role, tone, and reasoning approach for your specific use case. A persona applied to the chat input shapes every response until you switch or deactivate it.
Access pre-built prompts for common business scenarios — market analysis, strategic planning, data synthesis, and more. Click any prompt to auto-populate your chat input field.
Navigating to Smith
Smith is accessible from the main Compass navigation bar at the top of the screen. Click the Smith tab — it's the rightmost tab before Marketplace, and it displays with a distinctive icon when active. You'll land directly on the Custom tab within Smith, which is the main conversation interface. When you come back to Compass, whether you are reopening the browser tab or returning after working elsewhere, Smith opens on the landing page: the welcome state and the featured prompt cards, rather than the middle of a conversation you may have finished days ago.
Nothing is lost when this happens. Completed conversations stay in the chat history drawer, where you can search for one and pick it back up, and a conversation you were part-way through when you switched away is preserved rather than discarded. The change is only about where Smith puts you when you arrive: a clean starting point instead of an old thread you have to close before you can begin.
Smith is one of seven tabs in Compass. Your files live in the Lakehouse tab — Smith pulls from those files to generate context-aware responses. You don't manage folders or upload files inside Smith; that all happens in Lakehouse.
The Smith Interface
The Smith interface has four distinct areas. Understanding how they connect will help you get the most out of every session. You can resize the Smith panel inside Compass to suit the work in front of you. There is no separate fullscreen mode.
Three tabs sit at the top left of the Smith interface. CUSTOM is the main conversation area where you write prompts and receive responses. PROMPT LIBRARY shows pre-built prompts you can click to start a conversation. PERSONAS is where you create and manage role-based AI configurations.
The left panel shows your Lakehouse folder tree. Expand folders, check files or entire folders to include them in your query, and search by folder name using the search field at the top. You can resize the panel to give the file tree or the conversation more room. Files are processed for indexing as soon as they're uploaded and are usually available to query against shortly after — a background indexing cycle also runs as a catch-all to ensure everything is indexed.
The main conversation canvas. On the Custom tab homepage before a chat starts, you'll see Smith's welcome message and a selection of featured prompt cards you can click to begin. Once a conversation is active, this area shows the full message thread with real-time streaming responses.
The input bar at the bottom of the screen is your primary control center. It contains the text field for writing prompts, the Model dropdown, the Persona dropdown, a + icon for attaching files directly to the current conversation, a Settings gear icon, and the send button. Note: the + icon attaches files to the conversation only — uploading and organizing files into your Lakehouse happens in the Lakehouse tab.
Top-Right Controls
Two buttons appear in the top-right corner of the Smith interface, outside the tab bar. Start new chat opens a fresh conversation thread without clearing your history. Chat history opens a drawer listing your previous conversations. You can reopen any conversation from the drawer, and manage your conversations there.
Opens a new blank conversation. Your active model and persona selections carry over, but the conversation thread starts fresh.
Opens your conversation history drawer. Reopen any previous conversation, or manage your saved conversations from the drawer.
Choosing an AI Model
The Model dropdown in the chat input bar is where you select which AI engine powers your conversation. Smith supports multiple models simultaneously — each has different strengths in reasoning depth, context window size, creative generation, and response speed.
You have three options: select a single model, switch on multi-model and select several models for side-by-side comparison, or leave the selection blank and let Smith auto-select based on your query. Multi-model is off by default, so you turn it on when you want it. Available models include GPT, Claude, Gemini, and DeepSeek, with newer GPT and Claude generations, including the latest GPT releases and Claude Opus, Sonnet and Fable models, offered as advanced models. The list changes as new releases land, so check the Settings panel for what your project can use right now. Advanced models must be activated for your project before anyone can select them, and only activated models appear in the Model dropdown.
Best when you know which model suits the task. Gemini for large-context document work, GPT for complex reasoning, Claude for detailed analysis, DeepSeek for fast code generation.
Switch multi-model on, then select several models to see responses side by side. Useful for high-stakes queries where you want to evaluate multiple perspectives before acting on output.
Leave the model field blank and Smith picks the best-fit model for your prompt. A good default when you're not sure which model to use.
Additional advanced models are available beyond the core set, including newer GPT and Claude generations. They must be activated for your project before they appear in the Model dropdown, so check the Settings panel for what you can use right now. Smith's multi-model approach means no single model handles every task — the right choice depends on your specific query's demands.
Starting Your First Conversation
Smith works as a standard chatbot right away — no file setup required. You can start asking questions the moment you navigate to the Smith tab. File context is optional and can be added at any point.
Click the Model dropdown in the chat input bar and select an AI model, or leave it blank for auto-selection. Only the models your project has activated appear in the list.
Click the Persona dropdown next to the model selector and choose a persona to shape Smith's role and tone. Select Persona Select to leave Smith in its default, unframed mode.
In the My Lakehouse Files panel on the left, check the boxes next to individual files or entire folders you want Smith to reference. Files are processed for indexing shortly after upload. If a file row shows a yellow clock icon in the Lakehouse Type column, it is still being indexed and not yet queryable.
Type your question or instruction in the text field. You can also click any featured prompt card on the Custom tab homepage — it auto-populates the input field so you can start or edit before sending.
Click the send button. Smith streams its response in real time. If you no longer need a response that is still generating, you can cancel it while it is in progress. You can follow up with additional questions in the same thread — Smith maintains context across the full conversation, up to 30+ exchanges per thread.
Using Your Lakehouse Files in Smith
Selecting files from the My Lakehouse Files panel adds them to your query context. Smith can then reference, synthesize, and analyze those documents when generating responses — rather than relying solely on its general training knowledge.
You can select individual files by checking the box next to each file name, or check an entire folder to include all its contents. Multi-level folder selection is supported: selecting a parent folder automatically includes its subfolders and files. Deselect any item by unchecking its box.
Files held in the Lakehouse and files you upload directly to a conversation are kept separate, and attachment selection keeps the two apart, so it stays clear which set a file comes from. The same supported-format rules apply on every route, including attaching a file to a prompt with the + button. A folder cannot be attached directly to a Smith prompt; select folders from the My Lakehouse Files panel instead.
When a file is uploaded to the Lakehouse, it enters an indexing queue immediately. Most files are processed and available to query against in Smith shortly after upload. A background indexing cycle also runs as a catch-all to ensure any files not yet indexed are picked up.
A yellow clock icon in the Type column of the Lakehouse file list means the file is stored but still indexing. It is not yet a reliable source for questions in Smith, however clearly you have selected it. For most uploads this clears quickly, and once indexing finishes the normal file-type icon replaces the clock and the file is available to select.
The maximum size for a single file is 4GB, printed on the upload zone in the Lakehouse. The number of attachments allowed on a single Smith prompt is a separate limit, set by the platform and subject to change, so check the figure that applies at the point of upload rather than relying on a fixed number.
Document-Grounded Research
When you run extended research across documents you have selected, Smith is given the exact location of each file rather than working it out as it goes. In practice you will see far fewer cases where a research request reports that it could not resolve a document that is plainly sitting in your Lakehouse.
Nothing changes in how you set the work up. Select the files or folders you want in scope from the My Lakehouse Files panel and run your request as usual. The improvement is in reliability, not in the steps, so there is no new control to find.
Check the document's row in the Lakehouse before anything else. A yellow clock icon in the Type column means the file is stored but still indexing, and a file that is still indexing is not answerable however clearly it has been selected. Wait for the clock to be replaced by the normal file-type icon, then run the request again. Managing Files in the Lakehouse, linked below, covers the file list and its columns in full.
Supported File Types
Smith can work with the file formats the Lakehouse accepts. The types below are among those currently supported, and the same rules apply whether you browse for a file, drag it in, or attach it to a prompt. The upload area prints the current list and is the authoritative source:
What Smith Is — and Isn't
Smith is a decision-support tool, not a decision-making one. Its responses are generated by AI models and should be reviewed and validated before acting on them — particularly for financial, legal, or compliance-sensitive use cases. It surfaces patterns, synthesizes documents, and generates draft content; the judgment call is yours.
Smith does not use your uploaded documents or conversation content to train AI models. Your data remains within your organization's environment. Conversations are private by default — only authorized administrators with explicit audit access can review usage logs, and only within the scope of organizational governance requirements.
EBM Software employees cannot access your document contents or conversations by default. Any employee access requires explicit authorization for specific support, security, or compliance purposes, with a full audit trail of who accessed what, when, and why.
Learn how to use the Prompt Library and create personas that shape Smith's role and response style.
Manage file actions from within Smith, adjust Temperature and Memory settings, and access your conversation history.
Understand how the Lakehouse works, how files are organized, and how it connects to Smith.
Upload, organize, and manage the files that power your Smith conversations.
Smith works without any file setup — start with a simple question and add Lakehouse context when your work calls for it. The two are designed to complement each other, not require each other.
Comments
0 comments
Article is closed for comments.