Dev teams & standups
Sprint calls become searchable notes. Paste the transcript into Claude or ChatGPT and get action items, decisions, and changelogs without anyone typing.
[00:14:32] Priya Merge freeze starts Thursday — tag anything risky.
Discord voice → Markdown
Your voice channels, on the record.
Decibot joins the call, captures every speaker on their own track, and delivers a clean, chronological Markdown transcript — to your channel, your DMs, or straight into your AI's context.
Never guesses who spoke
Announces itself before recording
60 free minutes every month — no card
Who it's for
Wherever the conversation matters more than the call.
TTRPG parties
Every session, on the record. The GM keeps a full account of the campaign; the party gets "previously on…" recaps nobody had to write.
[02:41:07] GM The bridge gives way — everyone roll dexterity.
Communities & DAOs
Town halls, AMAs, and governance calls with a written record members can actually read — not a two-hour recording nobody replays.
[00:55:19] Mira Proposal 12 passes — treasury update goes live Friday.
How it works
Three commands between you and your transcript.
Invite it into the call
Decibot joins your voice channel and posts a public panel in chat, so everyone can see it's there and who's hosting. Or skip it — /start joins and starts in one go.
Start the record
Every speaker is captured on their own track from the moment they talk. Pause, resume, or restart the session anytime.
Stop & transcribe
OpenAI transcribes each track; Decibot re-times every utterance and stitches one chronological, speaker-tagged transcript.
Delivered where you want
Posted to the channel or DM'd to you, per /settings. Clean Markdown, readable by humans and ready for AI.
Also on board: /pause /resume /restart /cancel /status /account /settings — and every command is available to AI agents over MCP on Pro plans.
The difference
An audio file isn't a record.
Plenty of bots can record a call. What you want afterwards is not two hours of audio — it's what was said, by whom, in order.
They hand you audio
Multi-track recorders do one job: give you a big audio file. Finding one decision means scrubbing through the whole call at 2× speed.
standup.flac · 214 MB
They live somewhere else
Otter-style assistants are built for Zoom and Meet. Your community isn't leaving Discord — and a second app with a second login means nobody uses it.
another app, another login
Native to Discord, out comes the record
One slash command in the channel you already talk in. Out comes speaker-tagged, timestamped Markdown — the record itself, not homework.
standup-2026-07-02.md · 2 KB
Under the hood
Built like a good sound engineer works:
Built like a good sound engineer works:
capture clean, label everything.
Per-speaker capture
Each participant records on their own track, straight from Discord's voice gateway. Names are never guessed by an algorithm — the transcript knows who spoke because Discord does.
gpt-transcribe accuracy
OpenAI's gpt-transcribe reads each clean, crosstalk-free track — accents, jargon, and all — then utterances are re-timed to wall-clock and merged in order.
Markdown built for LLMs
Speaker tags, HH:MM:SS timestamps, zero filler. Paste it into Claude or ChatGPT and ask for the summary, the action items, or the recap.
Session controls
Pause, resume, restart, or cancel mid-session — from slash commands or the host panel. One host at a time per server keeps control unambiguous — the lock releases the moment a session ends.
Never loses a session
Kicked, dragged, or the channel empties? Decibot holds the audio it already captured and still delivers your transcript — even a mid-call crash can't eat a recording. Files are deleted once processing ends.
Agent access over MCP
Pro plans include an MCP server with full command parity — your AI agent can join a channel, run the recording, and get the transcript back as Markdown.
Consent, by design
Recording that announces itself.
Recording voice is a trust decision. Decibot is built so nobody in your server gets recorded without knowing about it.
Decibot is a verified app on Discord's App Directory. Details in our Privacy Policy and Terms of Service. Transcription is performed by OpenAI; your audio is never used to train models.
-
Visible presenceDecibot joins the voice channel as a normal participant — everyone in the call can see it, exactly like they'd see a person.
-
Public noticeA public panel posts in your text channel the moment a session begins, naming the host — no silent recordings.
-
Nothing keptAudio lives on disk only while it's being transcribed, then it's deleted. The transcript goes to you — not into a training set.
-
Minimum permissionsFive permissions, zero privileged access. Decibot can't read your messages or see your member list — Discord never even sends it that data.
-
Personal opt-outAnyone can run /ignore and Decibot will never capture their voice — in any server. Enforced before capture, not after.
-
Your rules, made easyRecording-consent laws vary by place. Decibot's visible presence and public notice make doing the right thing the default.
Pricing
Flat plans. Hard caps. No surprise bills.
Every plan includes full per-speaker transcription. When you reach your cap, recording pauses — it never bills past it.
Free tier
Add to Discord
Try it free — 60 minutes of transcription per server, every month. No card required. /start works the moment you add the bot.
Find your plan
Drag to your team's weekly hours in voice.
~17 h / month
Pro
$19.99/mo · $0.80 per hour
Starter
$9.99/mo
10 hours600 min of transcription / month
One team's weekly calls, fully on the record.
- Per-speaker transcription with timestamps
- All commands + host control panel
- Channel or DM delivery
- Hard cap — recording pauses at the limit
Pro
$19.99/mo
25 hours1,500 min of transcription / month
Daily standups, long sessions, and AI agent workflows.
- Everything in Starter
- MCP server access for AI agents
- Full command parity for agents
- 2.5× the hours at 2× the price
Enterprise
$49.99/mo
75 hours4,500 min of transcription / month
Big communities, long campaigns, heavy schedules.
- Everything in Pro
- Best rate — $0.67 per transcribed hour
- Priority support, direct from the founder
Need more? Overage packs: $5 per extra 5 hours (valid for current billing cycle)
Cancel anytime — access runs through the period's end
Check usage anytime with
/account
Payments handled by Stripe — your card details never touch our servers
FAQ
Questions, answered straight.
It doesn't guess — it doesn't have to. Decibot subscribes to each participant's audio track individually through Discord's voice gateway, so every word arrives already labeled with the speaker's name. There's no diarization model making probabilistic guesses, and no crosstalk bleeding between speakers.
Yes. Recording-consent laws vary by country and state — some require everyone's consent. Decibot helps by never being silent: it's visible in the voice channel like any participant, and it posts a public panel in chat when a session starts. And anyone who doesn't want to be recorded can run /ignore once — Decibot will never capture their voice, in any server. As the host, you're responsible for making sure your participants are okay with being recorded.
Audio is written to disk only while the session is being processed. It's transcribed by OpenAI, stitched into your transcript, and then the audio files are deleted. Your audio is not used to train models — by us or anyone else. Full details in the Privacy Policy.
Less than you'd expect. Audio is deleted after transcription, transcripts are delivered to you rather than warehoused, and Decibot never sees your text conversations — it runs on five permissions with no privileged access. What remains is your account and usage records, and those are yours to manage: export or delete everything yourself from your account page, instantly — no support email, no waiting.
Recording pauses with a friendly message — nothing is billed past your plan, ever. You can upgrade instantly, grab a $5 overage pack for 5 more hours, or wait for your period to reset. Hard caps are the point: your bill is the number you signed up for.
Yes — run /settings delivery dm and completed transcripts arrive as a direct message instead of posting in the channel. Handy for personal workflows, or for feeding transcripts to an AI without cluttering the server. Switch back anytime with /settings delivery channel.
Two ways. The simple one: every transcript is clean Markdown with speaker tags and timestamps — paste it into Claude or ChatGPT and ask for summaries, action items, or a session recap. The powerful one: Pro plans include an MCP (Model Context Protocol) server with full command parity, so an agent can join a channel, start and stop the recording, and receive the finished Markdown directly. Create an API key in your account and paste the ready-made config into Claude Code, Cursor, or any MCP client — the agent acts as you, in your servers, within your plan's hours.
Decibot uses OpenAI's latest transcription model, gpt-transcribe — built for real-world audio: accents, technical vocabulary, short phrases, and 50+ languages. Decibot's per-speaker capture compounds that: each track is free of crosstalk, which is where most transcription errors come from in group calls.
Yes — transcription covers 50+ languages, and the spoken language is detected automatically, so there's nothing to configure: run /start and speak. Accuracy is strongest in widely spoken languages; OpenAI publishes the supported list.
Get started
Put your next call on the record.
Add Decibot to your server, run /start, and read your first transcript moments after the call ends.
Decibot HQ is where questions get answered, features get requested, and launches get announced — come hang out even if you're just curious.