If you play D&D over Discord, someone at your table has already suggested the Craig bot. They were right to. Craig is the default recording bot for online tabletop, it earned that spot honestly, and if you searched “craig bot dnd” trying to decide whether to use it, my answer is yes. Use Craig. This post is about the question that shows up after you do, the one I built a whole product around: you now have six hours of audio. Who’s going to remember what’s in it?

What Craig gets right, and it’s a lot

Credit first, because Craig deserves it. It’s free. It’s open source. It sits in your voice channel for an entire marathon session without flinching. And its signature feature is genuinely great engineering: multitrack recording, each speaker captured on a separate track. If your cleric’s cat knocks the mic over mid-boss-fight, that chaos lives on one track, and you can pull it out without touching anyone else’s audio.

If what you want is audio, and there are excellent reasons to want audio, Craig is the right tool. Actual-play podcasters should use Craig. Anyone who edits their sessions should use Craig. A table that just wants a raw archive of game night, in case, should use Craig, tonight, for nothing.

We are not trying to out-record Craig. That race is over and Craig won it years ago.

Recording and remembering are different jobs

Here’s the thing I learned running my own games, and it’s the entire reason Loracle exists. A recording is write-only memory. Putting the session into a file is easy. Getting a specific fact back out of that file is brutal.

Play this out. It’s Wednesday. Your rogue messages the group chat: “wait, do we still owe the smuggler a favor, or did the barge thing cancel it?” Somewhere in your downloads folder is a perfect, lossless, multitrack answer to that question. It is buried at an unknown timestamp inside six hours of audio, spread across five tracks, mixed in with an hour of rules lookups and twenty minutes of everyone deciding on pizza. Nobody is going to scrub for it. So the table does what tables have always done: they argue from memory, the loudest memory wins, and the recording sits there, complete and useless, like a security camera nobody ever rewinds.

Every question that actually comes up between sessions is a question about content. What did the duke promise. What was the prophecy’s exact wording. Which name did the DM blurt out for that fisherman. Audio can’t answer content questions at conversation speed. Text can. That’s the whole gap.

The workaround everyone tries

I know what the resourceful DMs reading this are thinking, because it’s the obvious move and half of you have done it: download the Craig files, run them through a transcription tool, paste the transcript into ChatGPT, ask for a recap. And the first time, it works. You get a decent summary and feel like you’ve beaten the system.

Then the tax bill arrives, weekly. It’s a manual pipeline with four stops, and it runs at 11 p.m. when you’re the most tired person at the table. The transcript comes back as one anonymous wall of text, so “Speaker 3 agrees to the deal” is doing a lot of load-bearing work. And the paste-in has a hard ceiling: a chatbot can hold one session in its head, but your campaign is fifteen sessions and counting. The fisherman your party met in session three and the “old man on the dock” from session eleven are the same NPC. You know that. A fresh chat window with one transcript pasted in has no way to know that, and the moment continuity spans more than one paste, you’re the one stitching it back together. Congratulations: you automated the recording and became the pipeline.

What a recall engine does instead

Loracle starts where Craig stops, at the question “what’s in the recording?” It sits in your Discord voice channel like Craig does. You type /record at the top of the night and /end when you’re done. Then the machine does the clerical work: the session is transcribed with each speaker identified and labeled, in whatever language your table plays in, and a recap is posted to your channel before anyone has to remember whose turn it was to write one. Session after session, it builds the campaign’s memory: the NPCs, the places, the promises, connected across every session they appear in, into the kind of wiki that can actually settle a mid-scene dispute.

And when the smuggler argument breaks out on Wednesday, anyone at the table types /chat and asks. The answer comes back from the actual scene, the one with witnesses, in seconds. Nobody scrubs audio. Nobody re-argues it from vibes. I’ve written before about what the gap between sessions does to a table’s memory; this is the tool-shaped half of that answer. The human half, handing the recap round to your players, gets better when a real record exists, because the round can be loose and human while the record stays true.

The honest bottom line

Want pristine audio to edit, publish, or archive? Craig, no contest, and it costs nothing. Want your campaign to remember itself: who promised what, what the prophecy said, which NPC is secretly which? That’s a different job, it’s the one Loracle was built for, and a session recording that never becomes searchable text will never do it, no matter how beautifully it was captured.

Some tables will run both, the podcast track and the memory, side by side in the same voice channel. That’s a reasonable setup and I’d shake the hand of any DM running it. Just be clear about which problem each one solves before Wednesday’s argument starts, because “we recorded it” and “we can find it” have never been the same sentence.