Assistant

Assistant solves tasks across your project data. Context gives it capabilities, enabling it to understand your firm and project knowledge to deliver accurate, cited answers.

How Assistant Works

The Nomic agent searches, understands, and reasons over your project data to deliver accurate, cited answers. Powered by domain-specific models—Nomic Embeddings for semantic search and Parse models for file understanding—Assistant finds the information you need across files, drawings, and documentation. Give it context through integrations or direct uploads, and it will deliver answers grounded in your project data.

Assistant Controls

The input bar includes controls that adjust how Assistant processes your request:

• Model picker — Choose which model Assistant uses for the conversation. Different models offer different tradeoffs between speed and depth of reasoning. We default to models that give the best balance of speed and accuracy for most tasks.
• Deep Research — Toggle on to have Assistant perform a more thorough investigation across your files before answering. Deep Research takes longer but produces more comprehensive, well-sourced results — best suited for complex questions that span multiple documents.

AI usage limits

Your organization may enable per-user AI usage limits. When enabled:

• Assistant can show warning banners as you approach your limit.
• If your limit is reached, new AI requests are blocked.
• Contact your organization admin to increase your limit, or wait for the next billing-period reset.

Rendered Code Blocks

Assistant can render some code blocks as previews directly in the conversation. This is useful when a visual or interactive result is easier to review than source code alone:

• Diagrams — Mermaid code blocks render as interactive diagrams.
• Math — LaTeX and TeX expressions render as formatted math.
• HTML — HTML code blocks render with Preview and Code tabs. The Preview tab runs the HTML in a sandboxed frame, so interactive snippets can include scripts, styles, buttons, and small self-contained demos without access to your Nomic session, browser storage, or the surrounding app.

Use the Code tab to inspect or copy the underlying source. HTML previews work best when they are self-contained, with inline styles and scripts rather than external dependencies.

Context

Assistant's capabilities come from the context it has access to. Context is scoped per session and can come from two places: the project a session belongs to, and files attached directly to the session.

Project Context

When you start a conversation from within a project, Assistant automatically has access to all files and tags attached to that project. This includes files synced from external integrations like SharePoint, Egnyte, Autodesk Forma, and Bentley, as well as files uploaded directly.

Attaching Files to a Session

Use the Add menu in the Assistant input bar to attach additional files to the current session:

• Existing files — Browse and select files already in Nomic, including files from connected integrations and previously uploaded documents
• New uploads — Upload files directly from your computer for immediate use in the conversation
• Images — Attach images to give Assistant visual context, useful for reviewing drawings, photos, or any visual material

Example prompts (with your data)

Assistant works best when your question clearly uses project files or attachments from this session. Without that context, answers tend to be generic and not grounded in your data—for example, broad firm-wide research across projects or code checks without the right libraries configured may not match what you see on your first day.

Example prompts that align with typical uploads—drawings, manuals, specs, and submittals:

• “Summarize the sheet notes and general notes on the PDFs I attached.”
• “What maintenance clearances or access requirements does this operations and maintenance manual call out for [equipment]?”
• “Find every mention of [topic or product] in the spec sections I uploaded.”
• “Compare these two detail sheets—list clashes, missing references, or inconsistent dimensions.”

For code compliance, drawing standards review, or research across many projects, use the dedicated guides—Code compliance, Drawing Reviews, Active Project Research—and Workflows once those features are configured for your workspace. They package libraries, templates, and instructions so results stay consistent.

Permissions

Assistant can only access files that you have permission to view. If a project or folder contains files you don't have access to, Assistant will not see them or know they exist. This applies to files synced from external integrations like SharePoint and Egnyte — Assistant respects the same permission boundaries as the source system.

Tool Calls

Assistant has access to a set of tool calls that enable it to interact with your project data and solve tasks. These tools allow Assistant to search, read, and navigate your files and integrations:

Tool Call	Description
Search Files	Finds relevant files and content across your project data using Nomic Embeddings for semantic search.
Read Page	Extracts and understands content from files using Parse models to gather context for solving tasks.
Render Page	Directly places page pixels into a vision language model for visual understanding of drawings and documents.
List Directory	Explores available files and directories to locate the right information needed to solve your task.
Artifacts	Creates, edits, and reads artifacts — structured outputs like documents, tables, and drawing reviews that appear alongside the conversation.

Citations

Assistant grounds its answers in your project data and cites the sources it used. Citations appear as clickable references in Assistant's messages, linking back to the specific file and page where the information was found. This lets you verify answers directly against the original source material.

When Assistant uses multiple files to answer a question, each claim is cited independently — so you can trace every part of the response back to its source.

Conversations

Past conversations are listed in the sidebar and can be reopened or continued at any time. Conversations started from within a project also appear on the project page.

Session Visibility

Use the session actions menu to set who can view a conversation:

• Private — Only you can view the session
• Shared in Project — Project members can view the session (project sessions only)
• Shared in Org — Organization members with the link can view the session

Only the session owner can change visibility settings, rename, or delete the session.

To share a conversation, copy its link and send it to a teammate. Shared conversations are read-only, so recipients can view the full history but cannot edit or continue the session.

note

Files attached to a shared conversation are not visible to viewers who don't have access to them. However, any descriptions, quotes, or excerpts from those files that appear in Assistant's messages will still be visible in the conversation history.