Document Analysis

Upload files or attach URLs to a brief, then ask the AI to read, summarize, and extract information from them.

Overview

Document analysis lets you bring external information into a brief and work with it through the AI assistant. You upload files — PDFs, images, text documents, spreadsheets, code files — or paste in URLs, and they become part of the brief's context. From there you can ask the assistant to summarize them, answer questions based on what they contain, extract specific information, or incorporate their content directly into the brief document.

This is particularly useful for briefs that depend on reference material: research papers, design specs, meeting transcripts, third-party documentation, or anything your team has collected that should inform the work.

How It Works

  1. Attach documents to the brief context — From the Context tab in the brief sidebar, upload files or add a URL. Files are processed and stored as searchable context for the brief. Multiple files can be attached.

  2. Ask the assistant about them — With documents in context, ask the assistant to work with them. You can ask for a summary, search for something specific, extract action items, or ask the assistant to add content from a document into the brief.

  3. The assistant searches the context — When you reference your uploaded documents, the assistant performs a semantic search across the attached files to find the most relevant content. It returns answers with references to which document the information came from.

  4. Content is written into the brief — If you ask the assistant to incorporate findings into your document, it writes the extracted content into the appropriate section of the brief.

File Types Supported

The document analysis agent handles a range of file types:

  • PDFs — Full text extraction and analysis
  • Images — Visual analysis and description of what the image shows. Uploaded images receive an automatically generated plain-language description that is stored as alt text and used by all AI consumers (chat, plan generation, RAG). Hover an image in the Context tab to see its description.
  • Text files — Plain text, markdown, and similar formats
  • Code files — Source code with structure-aware analysis
  • Spreadsheets and data files — Tabular data and structured content
  • Video files — Analysis of video content
  • JSON and XML — Structured data formats

Interaction Patterns

You can work with attached documents in several ways through the chat panel:

Summarize: "Summarize my uploaded documents" or "Give me the key points from the files I attached." The assistant produces a structured summary organized by theme or document.

Search: "Find anything about authentication requirements in my documents" or "Look through the uploaded files for API rate limits." The assistant returns matching content with citations.

Answer questions: "Based on the spec I uploaded, what are the acceptance criteria?" or "What does the research document say about user preferences?" The assistant answers directly from the document evidence.

Analyze: "Compare the two spec documents I attached" or "What patterns appear across my uploaded files?" The assistant identifies themes, connections, and differences across multiple documents.

Extract: "Extract all action items from the meeting notes I uploaded" or "List the requirements from my attached spec as bullet points." The assistant produces structured output from the document content.

Update the brief: "Add the key findings from my uploaded research to the brief" or "Incorporate the decisions from the meeting notes." The assistant extracts the relevant content and writes it into the brief document.

In-Conversation File Uploads

You can also attach files directly in the chat panel without going through the Context tab. Attach a file to a message and ask the assistant about it in the same turn. The assistant reads the file and responds in the conversation.

When the file contains content that's relevant to the brief, the assistant can proactively offer to save it to the brief's context so it's available for future questions in that brief.

Key Capabilities

  • Semantic search: The assistant finds relevant content in your documents based on meaning, not just keyword matching. Asking about "error handling patterns" finds relevant content even if the document uses different phrasing.

  • Multi-document awareness: When you have multiple files attached, the assistant works across all of them and can identify connections, contradictions, or gaps between documents.

  • Source citations: Answers reference the specific document they drew from, so you know where information came from.

  • Direct document editing: Content from attached files can be written directly into the brief, formatted appropriately for a project document.

  • Persistent context: Documents attached to a brief remain available for the life of the brief. Any conversation in that brief can reference them without re-uploading.

Tips

  • Attach supporting documents before starting a brief writing session. The assistant will draw on them automatically when relevant, even without an explicit instruction.
  • For meeting transcripts, use the assistant's "process meeting notes" flow: it extracts decisions, action items, and discussion points and structures them into a brief section automatically.
  • If you have a large reference document (such as a lengthy spec or research report), ask the assistant to summarize it first, then ask follow-up questions on the summary rather than treating the full document as the target.
  • Use the Context tab to manage which documents are attached to a brief. You can review what's available and remove files that are no longer relevant to keep the context focused.
  • When attaching images — screenshots, diagrams, wireframes — you can ask the assistant to describe what it sees, identify UI elements, or compare a design to the requirements in your brief.

Related