typeheard
Drop a recording, get what was said.
typeheard transcribes the audio and video people already have (interviews,
lectures, podcasts, voice memos) with whisper.cpp
on our own hardware, and bills by the minute of audio instead of by the month.
- Free: the first 3 minutes of any file, no account.
- Whole files: one credit per audio minute. $5 buys 300 minutes, and minutes never expire.
- Agents: x402 per call, no account and no key.
- Formats: plain text, timestamped Markdown, SRT, VTT, JSON.
- Surfaces: web, HTTP API, CLI, TUI and an MCP server.
Built because people keep asking Reddit for "an Otter.ai alternative for files
I already have" and "transcription software for my thesis interviews".
Use it
curl -fsSL https://typeheard.pwa.pit.world/install.sh | sh # installs the typeheard CLI
typeheard interview.m4a # prints the text
typeheard talk.mp4 --format srt --out talk.srt
typeheard login # API key for whole files
typeheard tui
No install: npx -y @profullstack/typeheard interview.m4a (Node 22.6+).
curl -F [email protected] -H "Authorization: Bearer $TYPEHEARD_API_KEY" https://typeheard.pwa.pit.world/api/v1/transcripts
curl https://typeheard.pwa.pit.world/api/v1/transcripts/<id>/txt # also md, srt, vtt, json
MCP: npx -y @profullstack/typeheard-mcp with TYPEHEARD_API_KEY set. Its tools are
transcribe_file, get_transcript and list_transcripts.
How it is built
| |
|---|
apps/web | Bun + Hono: pages, API, magic-link + passkey auth, CoinPay top-ups, x402 |
packages/transcribe | ffmpeg + whisper.cpp via @profullstack/media2markdown-core; the formats |
packages/db | Postgres. transcripts is also the job queue (for update skip locked) |
packages/cli | @profullstack/typeheard: CLI, hqtui TUI, and the API client |
packages/mcp | @profullstack/typeheard-mcp: stdio MCP server |
packages/auth, payments, notify, config | shared with bg0ne.com |
A job is a row. The upload is written to DATA_DIR, the row is queued, and a worker
claims it, transcribes it and stores the segments. The upload is deleted as soon as the
words are out. Jobs survive a restart. A failure refunds the minutes it was charged.
Credit codes and invites
- Credit codes (
TH-XXXX-XXXX-XXXX-XXXX, from @profullstack/redeem-credits, stored only as
hashes) add transcription minutes to credit_ledger, once per account per code. Redeem on
/account ("Have a code?"), POST /api/v1/credits/redeem, typeheard redeem <code> or the MCP
tool redeem_code. Mint on the box:
docker exec typeheardcom-app-1 bun apps/web/src/admin/codes.js mint --minutes 300 --label "for Ada"
(list, revoke <id>, reconcile <user id>, quota <user id> <total>).
- Invites (
@profullstack/affiliates): three single-use links per account at /invites, each
/i/<code> for 30 days. The code rides the emailed sign-in link, and pays INVITE_BONUS_MINUTES
(default 30) to both people only when that link creates the account, the two are not one mailbox,
the new address is not a throwaway one, the inviter's account is a day old
(INVITE_MIN_INVITER_AGE_HOURS) and the inviter has had fewer than 10 bonuses in 30 days.
?ref= links are first touch, kept in the referral_code cookie and recorded in
aff_attributions when the link creates an account.
Run it yourself
docker compose up # Postgres + the app on :3000, payments off, everything free
Or bun install && DATABASE_URL=... bun run dev, with ffmpeg and whisper-cli plus a
ggml model where media2markdown-core looks ($XDG_DATA_HOME/media2markdown/).
| Variable | |
|---|
DATABASE_URL | required |
SITE_URL | public origin; the passkey rpID comes from it |
WHISPER_MODEL | path to a ggml model; the image bakes ggml-base.bin (multilingual) |
CONCURRENCY | transcriptions at once (default 1) |
PREVIEW_SECONDS | free preview length (default 180) |
RESEND_API_KEY, MAIL_FROM | sign-in mail |
COINPAY_API_KEY, COINPAY_BUSINESS_ID, COINPAY_WEBHOOK_SECRET | top-ups |
X402_ENABLED, X402_PAY_TO, COINPAY_X402_KEY | per-call payment for agents |
Secrets live in the logicsrc vault (typeheardcom--prod), not in a .env.
Test
DATABASE_URL=postgres://... bun test
MIT.
