You searched some version of “best way to record D&D sessions” or “best D&D recap tool,” and the results you got back are comparing tools that do completely different jobs. A recording bot, a transcription app, and a chatbot will all show up on the same listicle as if they were rivals, when they’re actually three stops on one assembly line. The reason your table still argues about the smuggler debt every Wednesday is that most tables build the first stop and quietly skip the other two.

I build one of the tools in this roundup, so you know where the road ends before we start walking it. But I’ve spent the last year writing honest, individual comparisons of each alternative, and this page exists to do what those listicles don’t: name the actual job, show you every way to get it done including the free way, and tell you plainly which kind of table should pick which.

The job has three parts, and most tools do one

When a DM says “I want to record our sessions,” what they want on Wednesday is an answer to a content question. What did the duke promise. Is the favor to the smuggler still live. What was the prophecy’s exact wording. Getting from a live session to that answer takes three distinct steps:

  1. Capture. Turn the night into audio without losing it.
  2. Transcription. Turn the audio into speaker-labeled text.
  3. Recall. Turn fifteen sessions of text into answers: recaps, a campaign record, and the ability to ask a question and get the scene back.

Every tool people recommend for this lives at exactly one of those stops. Once you see the assembly line, choosing gets easy.

Stop one: capture. Craig wins, and it’s free.

For raw recording on Discord, use Craig. It’s free, open source, battle-tested on marathon sessions, and its multitrack recording (every speaker on a separate track) is genuinely great engineering. If you want audio to archive, edit, or publish as a podcast, the race is over and Craig won it years ago. I wrote the full honest comparison in Craig Records. Who Remembers?, and the title gives away the catch: a recording is write-only memory. Putting the session into a file is easy. Getting a specific fact back out of six hours of audio is brutal, which is why the perfect recording usually sits in a downloads folder like a security camera nobody rewinds.

Verdict: if your goal ends at “we have the audio,” stop here, install Craig tonight, pay nothing.

Stop two: transcription. Otter is good at its job. Your table will stress-test its limits.

General transcription tools like Otter will turn your Craig files into clean, timestamped, speaker-labeled text, cheaply. For one session, that’s a real upgrade: searchable text beats scrubbable audio every time. The full breakdown is in Can You Just Use Otter for D&D?, and the short version of where it strains: a meeting transcriber knows that Mike spoke, but at your table Mike is also Kalden the ranger and half the NPCs in the scene, and no meeting tool has a slot for that. Fantasy proper nouns come back creatively mangled (Vex becomes “Beck’s”) unless someone maintains a custom glossary by hand, updated every time the DM invents a name mid-sentence.

Verdict: a fine choice for a table that wants transcripts, has a volunteer archivist, and runs short or episodic campaigns where cross-session questions are rare.

Stop three: recall. The DIY version is you.

The resourceful move, and half the DMs reading this have tried it: paste the transcript into ChatGPT and ask for a recap. The first time, it works, and it feels like you’ve beaten the system. I wrote up what happens after the first time: a four-hour transcript strains the context window, fantasy names drift between pastes, and a fresh chat has no way to know that the fisherman from session three and the old man on the dock from session eleven are the same NPC. The connections in a campaign live between the sessions, and a tool that sees one paste at a time never looks there. You end up stitching the continuity back together by hand, every week, which means you automated the recording and became the pipeline.

Verdict: works for a one-shot or a single catch-up. As a weekly system, the recall step stays a human job. Yours.

The free stack, assembled honestly

Put the three free-ish pieces together and you get the pipeline: Craig records, you download and feed the files to a transcription tool, you fix the names, you paste into a chatbot, you post the recap, you file the transcript, and you personally remain the cross-session memory. Every step is small. The sum is a part-time clerical job that runs at 11 p.m. on the night you’re the most tired person at the table, and it lands on the one person who already DMs.

I want to be straight about this: the free stack genuinely works for some tables. If someone at your table enjoys being the archivist, if your campaign is episodic, or if you’re a podcast table that edits everything anyway, build the free stack and spend your money on miniatures. You’ll be doing better than ninety percent of tables.

The purpose-built path

Loracle is my answer to the whole assembly line, so apply the appropriate discount to what follows. 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. The transcript comes back speaker-labeled, in whatever language your table plays in, and the recap is posted to your channel before anyone has to remember whose turn it was to write one. Mangled fantasy names get reconciled against the campaign’s own record instead of founding new characters each week.

The part the free stack structurally can’t do is the third stop: every session feeds a living record of the campaign, with the NPCs, debts, and prophecies connected across every session they touch. When the smuggler argument breaks out on Wednesday, anyone types /chat and the answer comes back from the actual scene, wording intact, in seconds. That’s the job. Everything upstream of it is plumbing.

Which table are you?

The podcast table. You edit and publish your sessions. Use Craig, full stop, and run anything else alongside it if you also want recaps. Multitrack audio is your raw material and nothing beats free.

The one-shot and short-arc table. Cross-session questions barely exist for you. Craig plus an occasional transcript, or even just a good spoken recap, covers it. Don’t buy a memory system for a campaign that fits in one.

The table with an archivist. Someone at your table genuinely likes the pipeline work. Bless them, build the free stack, and set a reminder for six weeks out asking whether they still like it.

The long-campaign table. Fifteen sessions and counting, promises from the spring detonating in the fall, and a group chat full of “wait, do we still owe the smuggler?” This is the table every general tool quietly fails, because the campaign layer is always, politely, your problem. It’s the table Loracle was built for, and the only one I’ll actually pitch: let the machine do the clerical work, and spend Wednesday arguing about what to do with the answer instead of what the answer was.

Whichever stack you pick, pick it on purpose. “We recorded it” and “we can find it” have never been the same sentence, and the best time to learn the difference is before the prophecy’s exact wording matters.