How to build an ELO DMS integration agent in Demodesk with folder routing

What and why

This guide walks you through building a custom AI Crew agent that takes your Demodesk meeting summary, finds the matching project folder in ELO DMS, and uploads the summary there — with a fallback folder for meetings that don't match a known project. The agent asks you to confirm the target folder before it uploads, so nothing lands in the wrong place. Use this pattern when you want to move Demodesk from an island tool into your existing document management workflow.

Who this is for

RevOps, IT, and operations leads at companies running ELO DMS (or any keyword-searchable DMS) who want every meeting summary archived in the right project folder without manual filing.

Prerequisites

• A Demodesk seat with AI Crew access (included on the Coaching & AI plan)
• An ELO DMS instance with API access and at least one service account with read/write permissions on the target folders
• A consistent folder-naming convention in ELO (project names, project numbers, or customer names that appear in meeting titles or transcripts)
• One catch-all fallback folder created in ELO for meetings that don't match a known project
• At least 3–5 historical meetings in your Demodesk workspace to test against

Steps

1. Open AI Crew and start a new agent — In the Demodesk app, open AI Crew from the Home sidebar and create a new agent. This opens the AI Agent Builder.
2. Describe the task in plain language— In the agent description field, write what you want the agent to do end-to-end. For example: “After every meeting summary is generated, search ELO DMS for a folder matching the meeting title or customer name. If you find a match, ask me to confirm before uploading the summary. If no match, upload to the fallback folder 'Unassigned Meetings'.” The builder uses this description to scaffold the steps.
3. Connect the ELO DMS integration— In the agent's tool list, add ELO DMS as a connection. If ELO isn't a native connection, use the Demodesk public API or a Make/Zapier scenario as the bridge. Authenticate with your ELO service account and grant the agent permission to search folders and upload documents.
4. Configure the keyword search step— Tell the agent which fields to pull from the Demodesk meeting (meeting title, attendees, customer name from CRM, key topics from the summary) and which ELO field to search against (folder name, project number, or metadata tag). Set the search to return the top match and a confidence score.
5. Add the conditional logic for the confirmation step— Build the branching rule:
   • If a folder match is found above your confidence threshold, the agent posts a message: “I found folder 'Project Müller GmbH'. Upload the summary here? (Yes / No / Choose different folder)”
   • If no match is found, the agent routes to your fallback folder or asks you to pick one manually.
6. Set the fallback folder— Hardcode the path to your “Unassigned Meetings” folder (or equivalent) as the default destination when keyword search returns nothing. Every summary gets archived, even when project context is missing.
7. Test the agent against historical meetings— Run the agent in your ELO test system, not production. Use 5–10 older meetings with known project assignments and check whether the agent finds the right folder. Tune the keyword logic and confidence threshold based on what it gets wrong.
8. Activate the agent and set the trigger— Once testing passes, set the agent to run automatically after every meeting summary is generated. It runs on every call from here without anyone touching it.

Tips

• Start with confirmation on, automate later.Keep the “ask before uploading” step in place for the first few weeks. Once you trust the matching accuracy, remove the confirmation for high-confidence matches and keep it only for ambiguous ones.
• Use CRM context, not just meeting titles.If your meeting titles are inconsistent (“Sync”, “Call”, “Quick chat”), pull the associated deal or account name from Salesforce, HubSpot, or Pipedrive and search ELO against that instead.
• Audit the fallback folder weekly. Anything landing there is a signal your keyword logic missed a real project. Use it to refine the search rules.
• Build one agent per workflow.If you also want to push summaries to your knowledge base or to Slack, build those as separate AI Crew agents. They're easier to debug and maintain.
• Test in a sandbox ELO instance. Never test routing logic against your production DMS. Older projects in a test system give you the same folder structure with no risk.

Related skills and agents

• AI Crew on Demodesk — the AI Agent Builder surface where you create custom agents
• Demodesk Marketplace — browse 30+ pre-built agents and 158+ skills
• Demodesk MCP server — connect Demodesk meeting data to other AI tools and workflows
• AI Sales Assistant — the agent that generates the summary your ELO agent will route
• Integrations overview — full list of native connections