Get answers from your files
Analyze selected workspace files to extract verified findings and generate a cited answer with clickable references.
Created by Chris Moen • Version 65 • 13 steps
Deep analysis for verified research
Finding specific information across multiple documents can be a slow process. This workflow automates the heavy lifting by scanning your workspace files to extract precise answers. Instead of skimming through PDFs or listening to meeting recordings, you get a direct response based solely on your data. This flow ensures your team bases decisions on facts rather than memory.
By using high-performance Large Language Models from OpenAI and Mistral, the system doesn't just summarise text. It cross-references information across different formats to ensure every claim is accurate and grounded in your original documents.
How the automated process works
The workflow follows a 13-step logic path to maintain data integrity. It begins by separating media files from text-based resources. If you have video or audio uploads, the system identifies and collects the associated transcripts. This ensures that spoken information is treated with the same weight as written documentation.
After verifying that the files are readable, the system prepares the input for analysis. It takes your original question and scans the content to produce a list of verified findings. To keep things organised, the flow assigns unique reference IDs to every piece of evidence. Finally, the Large Language Models synthesise these findings into a coherent answer. An automated audit step then checks the final response against the source material to prevent any errors before the answer is rendered for you to read.
Why use this research workflow
This template is ideal for research teams, legal departments, or project managers who need to verify facts quickly. It removes the risk of missing a key detail in a long transcript or a technical manual.
Key benefits of this automation:
- Clickable references: Every answer includes citations that link directly back to the source file for quick verification.
- Multi-format support: You can analyse text documents and media transcripts in a single request.
- Hallucination checks: The built-in audit step ensures the AI only uses information present in your files.
- Centralised knowledge: It turns your static workspace files into an interactive database.
- Consistent formatting: Every response follows a clear structure, making it easy to share findings with the wider team.
Steps
- Split media uploads from analyzable resources (function)
- Collect transcript resources from uploaded media (function)
- Account for media files without transcripts (function)
- Check at least one file can be analyzed (function)
- Prepare analysis input (function)
- Use original question (function)
- Flatten verified findings (function)
- Assign reference ids (function)
- Format findings for cited answer (function)
- Synthesize cited answer (llm)
- Audit answer for grounding (llm)
- Parse cited answer (function)
- Render answer with references (function)
FAQ
What does the "Get answers from your files" workflow do?
This workflow analyses your workspace documents and media to provide specific answers based only on your data. It extracts verified findings and provides clickable references so you can fact-check every claim immediately.
How does the automated analysis process work?
The system splits your files into analyzable chunks and converts media into transcripts. It then uses Large Language Models to synthesise an answer, which it audits for accuracy against your original files before displaying it with citations.
Which AI services does this workflow use?
This flow connects with OpenAI and Mistral to process your data and generate answers. It uses these services to synthesise information and perform grounding audits to ensure the responses remain truthful to your sources.
What is required to set up this file analysis flow?
You only need to select the files you want to analyse and provide your original question. The workflow handles the technical steps like flattening findings and assigning reference IDs automatically, so you can focus on the results.