Mem.ai built desktop recording in two days with Recall.ai

Executive Summary

Challenge

Mem wanted its desktop application to capture meetings and turn them into notes users could reference later.

The team initially built the recording infrastructure itself. What began with one engineer grew into a two-month project involving multiple engineers. Supporting Mac and Windows meant dealing with different drivers, permissions, audio devices, and poorly documented operating-system APIs.

The solution worked for most users, but the remaining edge cases generated support requests and weakened trust in a feature that needed to work every time.

Solution

Mem replaced its in-house infrastructure with Recall.ai’s Desktop Recording SDK.

The SDK handles permissions and audio capture across Mac and Windows while giving the team access to the raw audio and events needed for its own product. Mem sends that data through its existing transcription and AI systems to create live transcripts and organized meeting notes.

Within two days, the team rebuilt the infrastructure it had spent two months developing internally.

Results

With Recall.ai’s Desktop Recording SDK, Mem:

• Rebuilt two months of meeting capture infrastructure in two days
• Reduced related engineering effort by approximately 90%
• Reduced recording and transcription support volume by an order of magnitude
• Eliminated the long tail of recording-related support issues
• Supported reliable capture across Mac, Windows, meeting platforms, and audio devices
• Reached 50% weekly adoption for automatic meeting capture

How Mem uses the Desktop Recording SDK

Mem is an AI thought partner that remembers meetings, ideas, and research and brings that information back when it is useful.

When a user records a meeting through the desktop application, Recall.ai captures audio from the user’s microphone and the other participants through the outgoing app audio. The SDK also handles the required permissions on macOS and Windows.

Mem uses the resulting audio events to power waveforms and live transcription. When the meeting ends, the recording is sent to its backend, where its AI turns the conversation into an organized note with key takeaways and follow-up items.

“We use Recall.ai and specifically their Desktop Recording SDK to power our meeting recording experience within Mem,” said Kevin Moody, founder and CEO of Mem. “We use the Desktop SDK from Recall.ai behind the scenes to record, transcribe, and process that meeting.”

Two months of work was rebuilt in two days

The team initially expected building desktop recording from scratch to be straightforward. One engineer began building the connection, but the project quickly expanded as they encountered differences between operating systems and more edge cases.

After approximately two months, multiple engineers were working on infrastructure that was necessary but not central to what made Mem valuable.

“Our bread and butter at Mem is helping people organize and make use of that information later on,” Kevin said. “It’s not capturing the actual meeting audio recording.”

The team decided to test Recall.ai’s Desktop Recording SDK.

“Within two days, we had rebuilt all of the infrastructure that we had previously spent two months building,” Kevin said. “That was the moment where we realized we had to be using this instead of rolling our own solution.”

The SDK fit into Mem’s existing product

Mem already supported voice through its mobile, web, and desktop applications. It had its own systems for transcription, uploads, error handling, AI processing, and note generation.

The team did not want to replace that pipeline. It needed a reliable desktop capture layer that could plug into it.

Recall.ai gives Mem access to raw audio and event-based APIs. The team uses the SDK for permissions and recording, then sends the data through the same infrastructure used by its other applications.

“What was really nice about Recall.ai is that they give us access to some of the raw data,” said Scott Mackey, a founding engineer at Mem. “We’re able to take advantage of the permissions pieces and the recording pieces, but then pipe it right into the same stuff that our mobile app and our web app are both using.”

Reliable recording across Mac, Windows, and audio devices

Getting recording to work on one computer was not the difficult part. The challenge was making it work across operating systems, permissions, microphones, AirPods, audio drivers, and meeting platforms.

“Getting something working maybe just for one specific version of macOS is relatively straightforward,” Scott said. “Making it work across all the different platforms all the time, really reliably, ends up being really tricky.”

Mem’s in-house solution worked well for approximately 90% of users, but the remaining failures were difficult to reproduce and resolve. A meeting might appear to record correctly while missing one participant or audio source.

Since adopting Recall.ai, Kevin said those long-tail support issues have gone down to zero. Overall support volume related to meeting recording and transcription fell by an order of magnitude.

The time spent building and maintaining the desktop capture layer also decreased by approximately 90%.

“We’ve seen a 10x reduction in engineering effort required to build and launch these sorts of desktop-based features,” Kevin said.

Half of Mem users capture meetings every week

Automatic meeting notes had been one of Mem’s longest-standing feature requests.

The experience lets users stay focused on a conversation while the desktop application records it, transcribes it, and produces a note afterward. That captured information can then be organized and resurfaced elsewhere in Mem.

Following the launch of Mem 2.0, half of its users were using the product to capture meetings automatically each week.

“People have loved it above and beyond many of the other experiences within the product,” Kevin said. “Customers find it really intuitive, really reliable, and a great experience.”

More functionality without more capture infrastructure

Mem initially adopted Recall.ai to improve recording reliability, but the team also found functionality it had not yet built itself.

The Desktop Recording SDK can detect when a user enters or leaves a meeting, allowing Mem to prompt someone to begin recording or make the process more automatic. It can also connect each part of the transcript to the participant who said it, giving Mem names such as “Kevin” or “Sally” instead of generic diarization labels like “Speaker A” and “Speaker B.”

These capabilities make it easier for Mem to turn a recording into a useful, organized meeting note without building meeting detection and speaker identification in-house.

“The things we’re thinking about shipping next are already features in the SDK,” Kevin said.

By using Recall.ai, Mem can continue improving its meeting experience without maintaining the drivers, permissions, and edge cases underneath it.

“You could have your team try to build this themselves and iron out each edge case one by one, which will inevitably take weeks, if not months,” Kevin said. “Or you could go with Recall.ai and know that they’ve got your back.”

‍