Recording a Foundry VTT session when Foundry does not record

What Foundry’s built-in A/V actually is

Foundry ships "built-in technology to allow for audio/video (A/V) conferencing between all of the players in your game". Underneath, its documentation names the framework exactly: the WebRTC library simple-peer, which "relies on direct peer-to-peer connections to establish audio and video conferencing".

That phrase is the whole answer to "why can’t it record". Peer-to-peer means each player’s audio travels straight from their machine to each other player’s machine. There is no point in the system where the whole conversation exists as one stream that something could write to disk — not even on your own server, because the audio does not route through it. A recording feature would have to be built separately from the conferencing, not bolted onto it.

One requirement catches people setting this up for the first time: "SSL certification is a requirement for A/V Chat Integration due to limitations imposed by browsers." If your Foundry instance is served over plain HTTP, the browser will refuse to hand over a microphone at all, and the symptom looks like Foundry being broken rather than a certificate being missing.

The relay options do not change the recording answer

Foundry documents two third-party alternatives to direct peer-to-peer. Jitsi, where "all users connect to the Jitsi server instead of connecting directly to each other", which "often removes the necessity for port forwarding". And LiveKit, described as "a newer alternative to Jitsi", for when "the server host may not have sufficient bandwidth".

Both exist because peer-to-peer video is demanding — Foundry warns plainly that "video conferencing can consume a significant amount of bandwidth" and suggests looking elsewhere if your upload speed is low. Turning video off and running voice only is the cheapest fix, and for a game where everyone is looking at the map anyway it costs little.

What the relays do not give you is a recording. They change where the audio is routed, not whether Foundry captures it. Whichever mode you run, the capture has to happen somewhere else.

Option one: record the machine, not the app

The most reliable approach is to ignore Foundry entirely and record the computer’s audio output plus your microphone, with OBS Studio or any desktop recorder. It works regardless of which A/V mode Foundry is in, it survives Foundry updates, and it needs no module.

The cost is that you get one mixed track — everyone in a single file, no speaker separation. That is the same thing an in-person recorder produces, and it transcribes perfectly well; you simply will not get attribution. If you never needed to know who said which line, this is the answer and you can stop here.

Set the recorder to capture both system audio and your microphone, then test it for two minutes before session one. The classic failure is capturing only the microphone, so you record yourself narrating to silence for four hours and find out afterwards.

Option two: keep voice on Discord and let Foundry do the map

Plenty of Foundry tables never use its A/V at all. They sit in a Discord voice channel and use Foundry purely as the tabletop. If that is your setup, or if you are willing to make it your setup, the recording problem is already solved by someone else: a Discord recording bot gives you a separate track per speaker, which is the attribution the OBS route cannot provide.

This is the option worth choosing deliberately rather than falling into. Per-speaker tracks are the single biggest quality difference in what you can do with the audio afterwards, and it costs nothing but running voice in a different window. The setup is in the Craig guide, and the wider Discord workflow — including the options that are not Craig — is in the Discord recording guide.

It also sidesteps the bandwidth warning above entirely, since Foundry is then pushing map updates rather than video streams.

Option three: community modules, with a caveat

Foundry has a large module ecosystem, and modules exist that touch audio. Two honest warnings before you go looking.

First, Foundry’s own documentation does not describe any recording capability, so anything you find is a community project working around the peer-to-peer design described above rather than exposing something already there. Second, modules break across Foundry major versions, and a module that quietly stops recording is worse than no module at all — you find out when you go looking for the session and there is nothing there.

If you do use one, verify it after every Foundry update, and keep an OBS recording running underneath as a fallback for anything you cannot afford to lose. That belt-and-braces habit is standard practice for recorded games generally: two independent captures, because the failure mode of one is silence.

The part Foundry cannot help with at all

However you capture it, you end up with a large audio file of four hours of your table, and Foundry has no opinion about what to do with it. Transcription is a separate problem with its own trade-offs — free and local, cheap and metered, or done for you — and the honest comparison is in the transcription comparison.

The one Foundry-specific note: a fantasy campaign’s proper nouns are exactly what speech models get wrong, and Foundry tables tend to have more of them than most, because a group invested enough to self-host a VTT is usually invested enough to have named the tavern. Whatever you transcribe with, expect to correct names, and prefer anything that lets you supply your campaign’s vocabulary up front.