Turn any new file in a watched folder into ready to use study notes, briefings, and timelines. It is ideal for training teams and sales enablement where fast reading matters. It saves hours by turning long files into clear, useful documents.

New files start the flow with a local folder watcher. The file type is detected and text is pulled from PDF, DOCX, or TXT. The text is split into smaller parts, embedded with Mistral Cloud, and saved in a Qdrant collection. A retriever brings back the most relevant parts for each prompt. Mistral chat models ask and answer questions, then draft each template. An output parser and aggregator combine answers into final text, and a short wait helps avoid rate limits. The loop runs for every template you set, so you get multiple files from one source.

Setup is simple. Point the watcher to your folder, add Mistral Cloud and Qdrant credentials, and confirm the collection name. Expect a sharp drop in manual note taking and faster onboarding for staff who need quick summaries. Teams often cut prep time from two hours to under ten minutes per file. Use it for playbooks, research packs, customer files, and internal reports. Exports are saved next to the original with clear names.

• Watches a local folder and runs when a new file is added
• Detects file type and extracts text from PDF, DOCX, or TXT
• Splits long text into 2000 character chunks for better processing
• Creates embeddings with Mistral Cloud and writes them to a Qdrant collection
• Retrieves the most relevant chunks for prompts using a vector store retriever
• Loops through a list of templates to build study guides, briefings, and timelines
• Uses Mistral chat models to ask questions and draft final content
• Parses and aggregates model answers into a clean final document
• Exports generated files next to the source file with clear names
• Adds a short wait to help avoid model rate limits

• Reduce manual note creation from 2 hours to under 10 minutes per document
• Automate up to 90 percent of repetitive formatting and structuring
• Improve answer accuracy by using a vector store for context
• Handle 10 times more documents per day with the same team
• Connect Mistral Cloud and Qdrant to keep outputs consistent

1. Import the template into n8n: Create a new workflow in n8n > Click the three dots menu > Select 'Import from File' > Choose the downloaded JSON file.
2. You'll need accounts with Mistral Cloud and Qdrant. See the Tools Required section above for links to create accounts with these services.
3. In n8n, open the Mistral nodes and create a credential: double click the node, choose 'Credential to connect with' > 'Create new credential', then paste your Mistral Cloud API key from the Mistral API page. Name the credential so your team can recognize it.
4. Open the Qdrant Vector Store nodes and create a Qdrant credential: select 'Create new credential', then enter your Qdrant URL and API key from your Qdrant account. Confirm the collection name matches the node setting.
5. Open the Local File Trigger node and set the watch folder path. Keep polling on if your environment needs it. Make sure n8n has read access to that folder.
6. Check the Extract from PDF, DOCX, and TEXT nodes. Ensure they connect from the file type router so each file type is handled correctly.
7. Review the Recursive Character Text Splitter node and confirm the chunk size is 2000, which balances context and token cost.
8. Open the Get Doc Types node to edit or add templates. You can add more entries to generate different note styles.
9. Verify the Qdrant collection value in the vector store nodes. If you use a different collection name, update it in all Qdrant nodes.
10. Run a test: place a sample PDF or DOCX in the watched folder. Start the workflow and watch the execution. Confirm embeddings are created and stored.
11. Open your Qdrant dashboard and check that new points appear in the set collection. If not, recheck your API key and URL.
12. Check the Export to Folder node output. Confirm that generated files are saved next to the source file with the expected names. If you hit rate limits, adjust or keep the 2 second wait node enabled.