Skip to main content
Graphor’s Document Chat lets you have natural conversations with your documents. Ask questions in plain language and receive answers grounded in your actual content — with full conversational memory to maintain context across multiple questions. Document chat interface

​
Overview

Document Chat enables you to:
  • Ask questions naturally — No query syntax needed, just ask like you would ask a colleague
  • Get grounded answers — Responses are based on your actual document content
  • Maintain conversation context — Follow-up questions understand previous context
  • Scope to specific documents — Focus on one document or search across all sources
Document Chat uses the active parsing version of each document. The quality of answers directly depends on how well your documents were parsed. For better results, adjust the parsing settings and use advanced methods like MAI or Graphor for complex documents. See Data Ingestion for details.

​
Chat Options

Graphor provides two chat interfaces depending on your needs:

​
Chat with All Sources

Access the global chat from the Sources page by clicking the Chat button in the top-right corner. This mode searches across all ingested documents in your project, making it ideal for:
  • Finding information when you don’t know which document contains it
  • Comparing information across multiple sources
  • Getting a comprehensive view of your entire knowledge base
  • Answering questions that span multiple documents

​
Chat with a Specific Document

Access the focused chat from the Source details page (double-click on any source) by clicking the Chat button in the top-right corner. This mode searches only within the selected document, making it ideal for:
  • Deep-diving into a specific document’s content
  • Avoiding noise from other sources
  • Getting precise answers from a known document
  • Analyzing a single file in detail

​
How to Use Document Chat

​
Starting a Conversation

  1. Navigate to Sources (for all documents) or double-click a specific source (for single document)
  2. Click the Chat button in the top-right corner
  3. Type your question in the chat input
  4. Press Enter or click Send

​
Asking Effective Questions

Document Chat works best with clear, specific questions: Good questions:
  • “What are the main findings in the Q3 financial report?”
  • “Summarize the key terms of the service agreement”
  • “What products are mentioned and what are their prices?”
  • “Who are the parties involved in this contract?”
Questions that could be improved:
  • “Tell me everything” → Too broad, be more specific
  • “What?” → Provide context about what you’re looking for
  • “Info” → Ask a complete question

​
Following Up

Document Chat maintains conversational memory, so you can ask follow-up questions without repeating context:
You: What products are mentioned in the catalog? Chat: The catalog mentions three products: Product A (199),ProductB(199), Product B (299), and Product C ($149)… You: Which one is the most expensive? Chat: Based on the catalog, Product B is the most expensive at $299… You: What features does it have? Chat: Product B includes the following features: …

​
Starting Fresh

To start a new conversation without previous context, click the Clear button.

​
Use Cases

​
Research & Analysis

  • “What methodology was used in this study?”
  • “Summarize the literature review section”
  • “What are the limitations mentioned by the authors?”

​
Document Review

  • “What are the key obligations in this contract?”
  • “When does the agreement expire?”
  • “What are the payment terms?”

​
Knowledge Discovery

  • “What topics are covered in these documents?”
  • “Find all mentions of [specific term]”
  • “What recommendations are made?”

​
Quick Lookups

  • “What is the contact information listed?”
  • “What is the total amount on this invoice?”
  • “When was this document created?”

​
Best Practices

​
For Better Answers

  1. Be specific — The more specific your question, the more precise the answer
  2. Provide context — If asking about something specific, mention what you’re looking for
  3. Use follow-ups — Build on previous answers rather than asking unrelated questions
  4. Ask one thing at a time — Complex questions with multiple parts may get partial answers

​
Choosing the Right Scope

Use All Sources When…Use Specific Document When…
You don’t know which document has the answerYou know exactly which document to search
You want to compare across documentsYou want focused, precise answers
You’re exploring your knowledge baseYou’re analyzing a specific file
Questions span multiple topics/filesQuestions are about one document’s content

​
Parsing Quality Matters

Document Chat uses the active parsing version of each document to generate answers. The quality of responses directly depends on the parsing quality. Recommended parsing methods by document type:
Document TypeRecommended Method
Simple text documentsFast
Complex layoutsMAI or Graphor
Scanned documents / PDFs with imagesHi-Res or Hi-Res FT
Manuscripts / HandwrittenMAI
Multi-page tables, diagramsGraphor
If you’re getting poor answers:
  1. Double-click the document to open Source details
  2. Go to the Settings tab
  3. Select a more advanced parsing method
  4. Click Parse and wait for processing
The chat will automatically use the new active version for future questions. See Data Ingestion for detailed parsing method comparisons.

​
API Integration

Document Chat is available via the REST API: 
curl -X POST "https://sources.graphorlm.com/ask-sources" \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "question": "What are the main findings in this report?",
    "conversation_id": null,
    "reset": false,
    "file_names": null
  }'

​
Parameters

ParameterTypeDescription
questionstringYour question in natural language
conversation_idstring (optional)ID to continue a previous conversation
resetbooleanSet to true to start a fresh conversation
file_namesstring[] (optional)Restrict search to specific documents

​
Response

{
  "answer": "The main findings indicate that...",
  "conversation_id": "conv_abc123"
}
Use the returned conversation_id in subsequent requests to maintain conversation context.

​
Troubleshooting

If answers don’t match your questions:
  • Be more specific in your question
  • Use the single-document chat if you know which file has the answer
  • Check that documents are properly parsed
  • Try rephrasing your question
If the chat can’t find information you know exists:
  • Verify the document status is “Processed”
  • Check the active parsing version has the content
  • Try a different parsing method for better extraction
  • Ensure the content isn’t in images (may need OCR)
Response time depends on:
  • Number of documents being searched
  • Document size and complexity
  • Question complexity
For faster responses, use single-document chat when possible.
If the chat seems to forget previous context:
  • Don’t refresh the page mid-conversation
  • Use the same chat session for related questions
  • The API requires passing conversation_id to maintain context

​
Next Steps

Data Extraction

Extract structured data from your documents with custom schemas

Data Ingestion

Improve parsing quality for better chat responses

RAG Quickstart

Build your first RAG pipeline with Graphor

API Reference

Integrate Document Chat into your applications