What shipped in September: Correct what your agents suggest, and let them work with your files

You can now fix a Data Update Agent suggestion instead of throwing it away, and your AI tools can read and write the files attached to your deals. Those are September’s two headline releases. They both make agents easier to put to work. The first gives you more control over what the agent writes into Affinity, and the second lets documents move in and out of Affinity without a manual download or upload.

Two smaller releases follow the same theme. Each one removes a moment where you had to stop what you were doing, go somewhere else, and come back. In short, no more detours.

Edit Data Update Agent suggestions before they reach the record

A suggestion that is almost right shouldn’t cost you the same effort as one that is wrong. And until now, it did.

The Data Update Agent reads your meeting notes and proposes field updates. Previously, you had two choices for each suggestion: accept it or reject it. If the agent read a figure slightly wrong from a note, or inferred a pipeline stage that was close but not quite right, rejecting it meant retyping the update by hand. A near-miss was as useless as a miss.

Now you can fix a near-miss in place.

What's new:

• Edit in place. Click a suggested value and edit it directly, in an editor matched to that field's type, so a near-miss takes seconds to fix.
• The original stays visible. The agent's suggestion remains on screen next to your edit, and the row marks itself as edited, so you can see exactly what you changed before you apply it.
• A full trail. Activity History keeps both what the agent proposed and the value you applied, so every change is traceable.
• Mix and match within a batch. Edit some rows, accept others as they are, and reject the rest in the same pass, so one wrong value never holds up the rest of the batch.
• An honest edit history. If you edit a value back to the agent's original, it resolves as accepted, not edited, which keeps your edit history an accurate record of real corrections.

Nothing writes to Affinity without your approval, and approval now includes correction.

Where this helps:

• Post-meeting logging. Before, the agent would pull a revenue figure from your notes but read it slightly wrong. Now you correct the number before it reaches the record instead of rejecting the suggestion and entering it yourself.
• Pipeline hygiene. Before, the agent inferred a stage change that is close but not right. Now you adjust the stage rather than discarding the whole suggestion.

For relationship and entity fields, the editor lets you select records that already exist in Affinity.

The Granola integration will also allow Affinity to use Granola meeting notes to trigger the Data Update Agent in the same way that Notetaker notes do, so your records stay current whichever notetaker your team prefers.

Learn more in the Meeting Prep and Data Update help doc.

Move files in and out of Affinity from your AI tool

The documents attached to a deal are now part of what your agents can read, and what your agents produce can land back on the record.

Customers using Affinity MCP could already search records from Claude, ChatGPT, and other AI tools, but files stayed out of reach. Any workflow that involved analyzing a deck or attaching a generated document stopped at that boundary, because there was no way to get a file in or out.

What's new: Affinity MCP now supports file upload and download. These join the existing tools for searching files by keyword and retrieving files attached to a record, so an AI workflow can now start from a deal's documents and end with its output saved back on the record.

Try prompts like these:

• "Analyze the files we have on the Acme deal for mentions of 2024 revenue."
• "After generating the pre-meeting brief, upload it as a file to Affinity on this company record."

Where this helps:

• Diligence. Read a deck or teaser attached to an opportunity without downloading it first. Your AI tool pulls the file from Affinity and works on it where you already are.
• Brief distribution. Generate a pre-meeting brief and attach it to the company record, so the whole deal team can find it on the record.

Before you start: restart your AI tool and open a new session so the new tools show up. Files can be up to 100 MB each. See the supported file types for upload for the full list.

ICYMI: Build a market map in Claude or ChatGPT

A market map no longer has to start with a spreadsheet and a blank search.

Building one usually means screening companies against a thesis by hand, cross-referencing lists, and copying scores into a deck. And that work repeats every time the thesis gets revisited.

The Market Map skill in Affinity MCP lets you build one without leaving Claude or ChatGPT. Scoring draws on Affinity's own semantic search, so you start from the companies and data already in your CRM rather than a blank research pass. If the skill misses a company you already track, add it manually, so the map reflects your firm's full coverage.

Try this prompt: "Build a market map for [thesis or space]."

Learn more in the Affinity MCP Skills help doc.

Create records without leaving the form you're in

Meeting someone new should not mean abandoning the note you're halfway through writing.

Logging a note or interaction with someone who isn't in Affinity yet used to mean leaving the form to create the person somewhere else, then starting the note over.

Now, type a new name in the participants picker and an Add option appears inline. You can create a person, organization, or opportunity directly from the Log Interaction and Create Note forms, so the note stays intact and the new record is linked the moment you save. What you create persists even if you discard the note afterward, so you never recreate the same contact twice. Restricted opportunities keep their restricted flag and sharing limits, so confidential deals stay confidential.

How September's releases fit together

The two headline releases are the same idea approached from opposite directions. One gives you more control over what agents write into Affinity. The other gives agents more reach into what they can read from it.

The rest of the month cuts out the same kind of detour: create a record where you're already writing, and build a market map without leaving the AI tool you're already in.

Want to see how to put these agents to work for your firm? Book a demo or explore the Affinity help center for setup details.

Enjoyed this? More from this series

What shipped in August: New Lists is here, and it's not the only thing that got faster

What shipped in July: Ascend, and the infrastructure it runs alongside

What shipped in June: more Affinity data where your firm already works

FAQ

Can I edit a Data Update Agent suggestion before accepting it?

Yes. Click any suggested value to edit it in place, in an editor matched to the field type. The agent's original suggestion stays visible, and Activity History records both the original and the value you applied.

Can Affinity MCP upload and download files?

Yes. Affinity MCP now supports file upload and download, in addition to searching files by keyword and retrieving files attached to a record. Restart your AI tool and open a new session to access the new tools.

What is the file size limit for Affinity MCP uploads?

Each file can be up to 100 MB. Supported file types are listed in the Affinity help center.

Can I create a new person from the Log Interaction form?

Yes. Type a new name in the participants picker and select Add to create a person, organization, or opportunity without leaving the Log Interaction or Create Note form.