Alitu's loop is a chain of known steps: convert the recording, clean and level the voice, cut it by editing a transcript, add music, export an mp3 and publish an RSS feed. An agent can wire that together in a weekend from ffmpeg, a local Whisper model and a feed library. The weekend goes on the edges: noise reduction that sounds good on your voice, cuts that don't click, a transcript editor that stays in sync with the audio, and remote guest recording, which stays out of reach.
Before you buildAlitu has no free plan, but Audacity and a free podcast host cover recording, cleanup and a feed at no cost if you don't need cutting by transcript.
Build me a podcast episode maker to replace Alitu. Requirements: - Python 3.12 + FastAPI with a plain HTML and JavaScript front end; `uv run alitu` serves http://localhost:4810. ffmpeg 7 and ffprobe must be on PATH. - Storage: SQLite (stdlib sqlite3) at ./data/pod.db for shows, episodes (number, title, notes, status) and files; audio under ./data/audio/<episode>/ as original, cleaned and final. - Input: upload mp3, m4a, wav or mp4 and convert to 48 kHz mono WAV with ffmpeg, or record in the browser with MediaRecorder and upload on stop. Reject files over 2 GB. - Cleanup: one ffmpeg chain per file: highpass at 80 Hz, `arnndn` noise reduction with an RNNoise model from ./models, two-pass `loudnorm` to -16 LUFS, then `alimiter`. Show before and after with a player and the measured loudness. - Transcript: `faster-whisper` (model `small`, local) with word timestamps. The editor shows the words; selecting a span and pressing Delete cuts it, and a button cuts "um" and "uh". Render cuts with ffmpeg using 30 ms crossfades and keep the edit list so every cut can be undone. - Assembly: optional intro and outro from ./music, with the voice ducking the music (`sidechaincompress`). Export mp3 at 128 kbps with ID3 tags, cover art and typed-in chapters (`mutagen`). - Publish: write an RSS 2.0 feed with iTunes tags (`feedgen`) to ./site/feed.xml and copy the mp3s to ./site/ or to an S3-compatible bucket (`boto3`, keys in .env). Print the feed URL to submit. - Single user. No accounts, no telemetry, local-first; outbound calls only to the bucket. - Out of scope: remote guest recording, download statistics, a music library, video, social clips. - README: installing ffmpeg and the RNNoise model, bucket setup, submitting the feed to Apple Podcasts and Spotify, and a warning that every automatic cut and cleanup needs a listen before publishing.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Alitu.
Extend the podcast episode maker in this folder. Requirements: - A "Suggest chapters" button splits the transcript at pauses over 1.5 seconds and at topic shifts found by comparing sentence embeddings of neighbouring paragraphs (`sentence-transformers`, model all-MiniLM-L6-v2, local). - Show the proposed chapters with timestamps and one-line titles taken from the first sentence; let me rename, merge or delete them before export. - Write the result into the mp3 as ID3 CHAP and CTOC frames with `mutagen`. - Update the README with the limits of the method.
Add simple download statistics to the podcast episode maker in this folder. Requirements: - Serve the mp3s from a small redirect route `/dl/<episode>.mp3` that logs a row (time, episode, hashed IP plus user agent for the day) in SQLite and then redirects to the bucket URL. - Count one download per hashed visitor per episode per 24 hours and ignore known bot user agents; keep the list in a file I can edit. - A page `/stats` with downloads per episode and per week as inline SVG bars. - Update the README with the new enclosure URL in the feed and the privacy tradeoff.
Add a feed importer to the podcast episode maker in this folder. Requirements: - `alitu import-feed <rss-url>` downloads every enclosure (3 at a time, resumable), reads title, description, date, season, episode number and artwork with `feedparser`, and creates episodes marked published. - Keep each episode's original GUID in the new feed so directories don't show duplicates. - Print the lines to add on the old host for a redirect (itunes:new-feed-url) and do not modify anything remote. - Update the README with the redirect steps and a warning to keep the old feed up for weeks.
Alitu bundles the steps a new podcaster would otherwise stitch together from a recorder, an audio editor and a host, and it aims at people who never want to open an audio program. A script of your own removes the subscription but also the guide rails: you own the settings and the mistakes.
Estimate · Whisper and ffmpeg run on your machine. A static host or R2 bucket is free to a few dollars a month at podcast volumes. The build is token usage on your agent plan.
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Free trial | $0 | $0 | 7 days with all features, up to 2 hours of processed audio, and you can publish an episode. |
| Standard | $38 | $32 | All features: recording, editing, audio cleanup, music library and hosting. Unlimited episodes and storage under a fair-usage policy, 12 hours of transcription per billing period. |
| Two shows | $63 | — | Standard for two shows. |
| Done-for-you editing | $295 | — | Weekly or fortnightly editing by Alitu's team with a 4-day turnaround. |
No free plan; a 7-day trial lets you publish an episode and, per the pricing page, process up to 2 hours of audio.
Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: Recording remote guests and calls inside the app, Hosting with download statistics and help submitting to directories.
About $38 a month on the Standard plan, $456 a year (checked 2026-10-19).
Recording remote guests and calls inside the app; Hosting with download statistics and help submitting to directories; The built-in royalty-free music library; The polish of a tested cleanup chain and transcript editor; Chapter and social-clip tools, and the transcript used for show notes. If any of those matter to you, keep paying.
Yes: Audacity (open source), Audapolis (open source), Castopod (open source).