Payment error spike · 41 in 5 minutes · View run
Your accumulated conversations are a company asset, but you can't find them when you need them. Modak AI reads our conversations for us, finds things by meaning, and does work on schedule. There's nothing extra to install and no new accounts to open. Just talk to it in the messenger you already use.
The answer stays in the channel, so the whole team learns it together. You asked one line, and it even checks "has this come up before?" without being told. Call it inside a thread and it answers in the thread.
@Modak We were going to A/B test the landing page copy last week — did we reach a conclusion?
Yes, it was settled on 08-21 as option B.
I searched all of our conversations and found no post saying it was deployed. There's only a request in #design and no reply. Seo-jun Lee is in charge — want me to ask him?
Just ask in a full sentence in the search bar. It searches our conversations to answer and attaches a link to the original message. Only you see it, and it doesn't stay in the channel.
Confirmed at ₩12,000 per person. It was settled in #sales on July 18, and nothing has changed since. There was one "let's revisit at year-end" comment on August 2, though.
Ask about "the day the server blew up" and it still finds posts that say "outage," "502" and "recovery." Every conversation is indexed by meaning. Ask, calling from a channel, and agents all use this search.
Missed-mention guardian, morning briefing, unanswered-question patrol, customer-inquiry watch, weekly report draft… Write it in plain words and Modak fills in when and where to look. If something's wrong, a person can fix it. If there's nothing to report, it stays quiet.
Dev bot, Ops bot, Research bot, Strategy bot, Office bot, CS bot, Data bot, Marketing bot, HR bot, Finance bot, Legal bot, Assistant bot. Each one works in its own isolated workspace: it creates files, runs code, and digs through materials. They use our conversations and company data as their basis, and build up memory as they work. Each role has set models, and in most cases the smart router picks one to fit the request. Modak called in channels and conversation search run only on models on servers in Korea (see what is in Korea).
@DevBot Find the cause of the payment error spike and open a PR
The cause was a missing retry in the refund API. I opened PR #128 and all 14 tests pass. Merge it after the review bot's review.
PostgreSQL, MySQL, Google Sheets, Notion, Confluence, Jira, GitHub, GitLab, REST API, MCP. Read-only, never write. Runs a single statement and cuts it off within 20 seconds. Passwords are stored only as ciphertext, and internal network addresses are flatly rejected.
SELECT count(*) FROM users WHERE created_at ≥ '2026-09-01'Read-onlyFor facts that can't be answered from company conversations alone, it checks outside. It reads what the question means, picks and attaches only the data needed, and answers with the values and sources it verified. It never makes up numbers by guessing. If a source is down, it says so.
Every message has a translate button. 15 languages into your screen language, visible only to you. Mentions, channels, code and numbers stay as in the original. Polish in the input box only fixes wording and never changes content. Undo with ⌘Z.
Có thể gửi lại bản hợp đồng cho @Park Jihoon trước 3 giờ chiều không?
Incoming webhooks, a Slack-style API (chat.postMessage and more), and event subscriptions. You don't need to fix the notification scripts you used with Slack. Modak AI reads posts made by bots and uses them as sources. Building one takes 3 steps and 1 minute. After logging in, create one under "Modak AI › Connections & bots › Create a bot", or open it from "Bots & webhooks" in the ＋ Create menu on the left rail. Any member can create one, on any plan.
POST https://modak.im/hooks/•••••Slack-style /services/T…/B…/… works as is, too@Modak Did that alert we just saw happen last week too?
Yes, the same alert fired twice in the early hours of August 29. The cause then was missed retries. That day's thread
Set the channels, repositories, and webhooks to watch, and it checks every 2 minutes. If the same problem repeats or is serious, the autopilot bot creates a branch and fixes it, the review bot reviews it independently, and if tests pass it merges and deploys. You decide in the wizard how far it goes automatically and whether human approval is required.
See exactly what agents and automations did, when, at whose request, which model was chosen, how many seconds it took, and what it cost. No metering is hidden.
| When | Who · What | Model | Time | KRW |
|---|---|---|---|---|
| 09-07 08:30 | Morning briefing · Jihoon Park | Qwen Flash | 4.1s | 6 |
| 09-07 13:21 | @ResearchBot · Jihoon Park | Smart router | 48s | 31 |
| 09-07 16:09 | @DevBot · Hangyeol Seo | Claude Sonnet 5 | 6 min 12 s | 204 |
| 09-07 17:40 | Missed-mention keeper · Seojun Lee | Qwen Flash | 1.9s | 4 |
| This month · 214 times | 3,912 | |||
If it's an AI-native messenger, the door should be open for outside AI to come in. Put one standard connection (MCP) address into your AI app and log in, and that AI reads channels, summarizes, replies in threads, and checks DMs within your permissions. It's free regardless of plan.
Modak, summarize today's discussion in #marketing and reply in the thread with only what was decided.
#marketing I read 23 posts and 4 threads today. Two decisions: ① lock in landing page B ② send the newsletter Friday at 10. I left a two-line reply in the "Just tell me the decision" thread. View reply
These are the first things to ask when AI joins your team. We've locked every one of them in as a rule.
Private channels and DMs the caller can't see, Modak can't see either. It doesn't take over permissions.
Every answer comes with a link to the original. It makes no claims without a link.
If it can't find something, it doesn't make it up. It says "couldn't find it" and tells you how to search again.
It never writes to or deletes from the databases and documents you connect. One sentence at a time, within 20 seconds, and internal network addresses are refused.
If there's nothing to report, it says nothing. No notification fatigue builds up.
It is stated in the privacy policy. Modak called in channels runs only on models on servers in Korea, and the exceptions are published in a table.
That's one third of what Slack Pro charges each person every month (US$7.25 · about ₩9,700), and one quarter if billed annually. We only charge what it costs to run the models.
Each time you pay, we calculate by the headcount at that moment. Even if payment stops, your conversations and files stay as they are.
We calculate based on the number of members at the time of each payment.
Credits grow with your headcount.
Slack is a registered trademark of Slack Technologies, LLC, and Jandi, KakaoWork and Naver Works are trademarks of their respective companies. Modak is not affiliated with any of them. Slack prices are based on the official slack.com price list as of September 7, 2026 (US dollars, excluding VAT) and may change afterward. Based on an exchange rate of ₩1,344.
"Missed-mention guardian" alone is enough. It turns on in 30 seconds, and from that day nothing slips past you.
Use it without registering a card · even the free plan gives each person 30 uses a month