Getting started
How do I auto-edit my podcast with AI?
Drop the episode in and about a minute later Cast plays you 30 seconds of YOUR recording already edited — long pauses tightened, filler words cut, the voice cleaned up, music sitting under it. That is the first pass. You hear the result before doing any work, you see every cut in the text, and nothing touches your episode until you say so.
Every other editor asks for the work first and the verdict second: import, cut, clean, score, and only then find out whether it was worth the evening. The first pass turns that around. The result arrives first; the work is what you do afterwards, if you want to change something.
It is your audio, not a sample reel. Cast takes the opening minute of the episode, edits about 30 seconds of it for real, and plays that back. Everything you hear was decided on your recording: which pauses were too long, which "um" got in the way, how loud the music should sit under your voice.
Nothing is committed. The preview is a proposal — close it and your episode is exactly as you uploaded it.
Want to hear it before uploading anything? The Cast home page has four before / after examples: raw takes with the usual problems, and what the first pass made of them. One button plays the raw lines, then the same lines after Cast.
Automate the edit, keep the control. The pass does the pass; the cuts, the music and the final call stay yours to change. No result you did not hear first, and no credits spent on something you did not ask for. Drop an episode and listen.
- A/B against the raw takeSwitch between With music, Voice only and Original on the same passage. The raw version is right there, so the difference is a click, not a memory.
- Every cut is visibleThe transcript under the player marks what was removed and why — a long pause, a filler, a false start. Nothing happens off-screen.
- Reversible, all of itApply it to the whole episode and every change lands as one undo. The Pauses and Fillers panels still list each cut, so you can restore any of them one by one.
- Music that knows it is a bedThe bed is picked to sit under speech and ducks under your voice automatically — you hear the mix, not a track fighting the dialogue.
What Cast decides for you in that minute
Long pauses are shortened to a natural beat rather than removed, so the conversation keeps its rhythm. Filler words and false starts come out where they interrupt the sentence. The voice is cleaned and levelled to broadcast loudness. A music bed is chosen for the kind of speech this is and mixed under the voice with ducking on.
The preview also names the episode: the title you see in the dialog is written from what you actually said in that first minute.
From the preview to the whole episode
"Apply to the whole episode" runs the same pass end to end: every pause, filler and fix across the full recording, plus chapters, plus the voice clean-up, plus a full-length version of the bed. It is priced per started 10-minute block — 100 credits each, a block counting once the episode is a minute into it — so a 30-minute episode costs the same whether you apply it now or after an hour of manual edits.
Prefer to drive yourself? "Edit it myself" closes the dialog and leaves the transcript untouched — the first pass is a suggestion you can ignore.
Automated, but never out of your hands
Automating the edit is worth nothing if you cannot see what was automated. Every cut the pass makes is an op on a word, not a re-rendered file: it shows struck through in the transcript, it is listed in the Pauses and Fillers panels, and it comes back with a click. Applying the pass to the whole episode lands as one undo, and a checkpoint is written before it, so the version you had is one restore away.
The spending is the same story. The first pass on an account is free and runs by itself. After that nothing is ever debited silently: the card offers the run, the price is on the button before you press it, and a run that fails is refunded. Applying to the whole episode shows its price too — 100 credits per started 10 minutes.
And if the pass gets it wrong, that costs you a click, not an episode: close it and your recording is exactly as you uploaded it.
Or hand it to your own AI agent
The same pass is available over MCP, so Claude, Cursor, n8n or a script of your own can run it instead of you clicking. One command connects it — claude mcp add --transport http cast https://cast-mcp.mubert.com/mcp, then you approve it once in the browser — and then you ask in plain language: "produce an episode from this recording, tight interview, light bed".
The control promise holds there too, and it has to: an agent spends faster than a person. Each connection carries a "can spend credits" switch that is off by default, and even with it on every paid step quotes its price first and only spends when the agent confirms that exact quote. The agent's pass opens in Cast as ordinary work — the cuts land on the timeline and as word-anchored edits in the transcript — so you can let it do the boring part and finish by hand.
Start with connecting an agent, then see what it can and cannot spend.
How long it takes
Under a minute on a short take; on a long episode the wait is the transcription behind it. The screen names the step it is on — reading your episode, writing the transcript, finding pauses and fillers, editing 30 seconds with music — and you can leave for the dashboard at any point; Cast tells you when the pass is ready.
On the free plan you get one pass a month, and the first one on a new account is free on top of that. Nothing else is gated: listening, comparing and editing by hand stay free whatever your plan.
Last updated