Papermark Agents - Let AI agents run your next deal, fundraise or data room

So good even unicorns are trying to copy you

We've used Papermark before for document sharing, and the data room experience was fine but still required a lot of manual follow-up. Having an AI agent handle investor questions or track who's viewed what during a fundraise would save real time. I'm curious how much human oversight is built in — would founders still review what the agent sends before it goes out, or is it fully autonomous?

Going from 2–3 hours of manual M&A data room setup to a single agent prompt is genuinely compelling. Curious how you handle the audit trail when the agent mints watermarked links — does the log capture which human authorized the agent action?

The document chase in a B2B deal or fundraise is almost always where momentum stalls. Someone shares a data room, the other side goes quiet for three days, then comes back with a list of what is still missing. An agent that tracks what has been uploaded, reviewed, and flagged incomplete feels like it could cut that dead time significantly.

I am curious whether Papermark Agents can detect gaps proactively. Like if a deal partner expects to see specific document types and they are missing from the room, can the agent flag that before the review meeting rather than after it?

Giving an agent 40+ MCP tools over a data room is powerful, but 'create a tracked external share link to this confidential deck' is exactly the action I'd want gated. Can an agent token be scoped to specific data rooms or to read-only/analytics, so a misfired prompt can't mint an external-shareable link to the wrong document? And in self-hosted mode, do those MCP tools run fully against my own instance, or does link-tracking still route through Papermark's cloud?

I've definitely had moments where I wondered, "Did they actually read the document or just download it?" 😄 It's nice to see a tool solving that uncertainty.

open source data room + agents is smart. congrats!! 🎉

Can you set an approval step before an agent generates or sends any investor link?

The agent handling investor Q&A at 2am while you sleep is the actual selling point. That loop — request in, docs surfaced, access granted — is something you need motion to feel. Congrats on the launch.

The permission-scoping thread is well covered, so a different worry from building agent surfaces: 40+ MCP tools is a lot for a model to route across reliably. Past roughly 20 tools I have watched agents mis-select, reach for the analytics tool when they meant to mint a link, or stall mid-plan. Did you group these into a few high-level intents like setup_room or share_with, or expose all 40 raw and let the model sort it out? Curious which held up in real testing.