Everything you want to know about MLOG
From product and languages to privacy and billing — and if you can’t find your answer, just write to us.
Product & features
No. MLOG sends no bot and never joins the call — it captures this computer’s system audio and mic on two separate channels. The other side sees no “meeting bot”.
No. It’s a quiet sidebar floating beside your call, and it only ever adds to that sidebar: it never pops, blinks, auto-expands or makes a sound, and you can collapse it anytime. Two things land there. When the other side uses a term that already appears in your own glossary or uploaded documents, it is added silently, with the source, and at no cost — it is looked up in your material, not generated. And when you ask, you get one answer, one data point and an evidence link.
Every conclusion carries a timecode link to the exact words and audio. When it can’t find a basis, it says “not found” — it never produces pretty but unaccountable content.
No. Outbound emails are only saved as drafts and go out solely after your explicit confirmation. The system never sends on its own — a human is always in the loop.
No. Audio is encrypted to this computer’s disk before any cloud processing (local-first). Recording continues through a network outage; once you’re back online it uploads and catches up the transcript. A crash loses ≤5 seconds — we treat any failed capture as an incident.
Due to operating-system and third-party app policy restrictions, we do not promise reliable capture of online-meeting audio on mobile. Mobile is built for in-person and on-the-go scenarios — offline capture, briefs, Advisor Q&A and post-meeting confirmation all happen on your phone. For online meetings, capture on desktop.
Languages & transcription
The interface and outputs support 8 languages: English (en), Español (es), Deutsch (de), Français (fr), 日本語 (ja), Português (pt), 简体中文 (zh) and 한국어 (ko). Switch anytime in Settings.
In-sentence code-switching is MLOG’s flagship capability, and you never declare a meeting’s language up front: the transcript keeps every language exactly as spoken instead of folding them into one. Search and minute generation understand the mixed context too. Chinese-English is the pair we have tested most thoroughly; other languages are labeled by tested engine capability.
Yes. A minute in your team’s language and a follow-up email in your client’s language are generated together as a pair; the language pair is set per meeting or per workspace. Every decision still carries its timecode link.
Privacy & security
Yes. Audio is encrypted to this computer’s disk before processing, encrypted in transit and at rest, and a crash loses ≤5 seconds. The database is in Singapore and recorded audio in the Asia-Pacific region; your data is never used to train any model, and our AI vendors are bound by the API terms under which we use them.
In the cloud, the database and its backups are in Singapore and recorded audio is in object storage in the Asia-Pacific region; speech recognition, the AI models and search indexing run at our vendors, in the United States or globally. The desktop app keeps a local copy of your original recordings on your own device, up to the on-device cache limit you set; the mobile app and browser capture release each local segment as soon as it has uploaded. See the Privacy Policy for details.
Never. Your recordings, transcripts and minutes never enter any model training — not ours, and not that of our speech-recognition, language-model or search-indexing vendors, who are bound by the API terms under which we use them.
You can export or delete all of your data from Settings at any time — the access, portability, correction and erasure rights granted by GDPR / CCPA. Deletion completes across our production systems within 24 hours, with a confirmation sent to you; encrypted database backups are held for a limited period and are overwritten on a rolling basis.
Security & compliance
Data takes exactly three steps, encrypted at every one: first AES-256 encrypted to your own disk (offline-safe, ≤5s loss on crash); TLS 1.2+ in transit; then encrypted at rest, with the database in Singapore, the object store in the Asia-Pacific region and no cross-region replication configured. Deletion cascades.
Four promises built into the product’s behavior: your data is never used to train any model; meeting content is workspace-only by default; outbound sends always need human confirmation; deletion cascades everywhere (audio, transcript, minutes, index and share links). Full text in the Privacy Policy.
Four categories touch your meeting content, each under least privilege: cloud infrastructure (database in Singapore, object storage in the Asia-Pacific region), speech-recognition providers, LLM providers and search-indexing providers. The last three process it in the United States or globally and, bound by the API terms under which we use them, do not train on it. The categories are listed in the Privacy Policy appendix; write to us and we will send you the current named list.
Data-subject rights — access, rectification, export and erasure — can be started in the app and are answered within 30 days; we sell no personal information, and we create no voiceprints and process no biometric data. If your organisation needs a data processing agreement or Standard Contractual Clauses, how to arrange them is set out in the Privacy Policy.
Billing & subscription
There is no free plan. All three tiers bill monthly and are priced in USD: Starter and Plus (hard-locked at 1 seat each) and Team (3 seats included). All three place no limit on the number of meetings; each tier comes with a pool of credits that four services draw from at different rates, and each meeting draws its actual duration — never rounded up. Subscriptions belong to the workspace — see the pricing page for allowances and prices.
Nothing stops, and we never buy credits for you automatically. As credits run low, services shut off in order of how safely they can wait — the live Advisor first, then cross-meeting questions, then cloud OCR — and recording and transcription are always the last to stop. Local capture and writing to disk never stop at all. When credits reach zero, cloud post-processing is held, not failed: add credits or wait for the next period to reset and it finishes on its own, with the held audio kept at least 30 days. Buying more takes one deliberate click, and the price is written there before you click. We will not debit you first and tell you afterwards. Usage is itemised and auditable.
Credits are the only meter in the product. One sentence is enough: 1 credit = 1 minute of recording, live captions run at the same rate, and the live Advisor is 2 credits per question. Recording and transcription costs 1 credit per minute and covers desktop capture → transcription → bilingual minutes, plus any historical recordings you upload. The live in-meeting Advisor costs 2 credits per question — it is our most expensive item, so the rate is stated plainly rather than hidden. Terms the Advisor surfaces from your own glossary and documents while you talk are free: they are looked up in your material rather than generated, so there is no question to charge for. Cross-meeting questions are 5 credits each and cloud OCR fallback is 1 credit per page; on-device OCR in the desktop app is free and unmetered. Credits included with your plan are granted monthly and do not roll over; pack credits stay valid for 12 months if bought on Google Play and do not expire if bought through the App Store. Multi-seat plans pool them across the workspace. See the pricing page for the per-plan allowances.
Because what Team sells is “a team’s meetings become an organization you can query”, and organizational memory only means anything when it spans people and seats — a one-seat Team plan would be a contradiction. At 3 seats that is $37.33 per seat. That per-seat number lands squarely inside the price band of comparable team plans — and nobody in that band gives you team-wide live in-meeting advice. If your team needs more than 3 seats, write to support@mlog.ai.
Yes — and that is the design, not a hole we forgot to close. A seat is a recording seat: it buys the right to have the meetings you personally attend captured, transcribed and analysed. The people reading the minutes are Guests: free, unlimited and never downgraded, scoped to the single meeting explicitly shared with them. The minutes already exist; one more reader adds no recognition or inference cost. Gong, Avoma, Granola, Circleback and tl;dv all do the same — charging readers is the strange choice.
No — and this is not a policy rule, it is physics. Desktop capture records only the audio your own machine can hear: to record a meeting, you have to actually be in it. One seat is one calendar, and one person’s time has an upper bound; concurrent capture is also 1 stream per seat, so you cannot be in two meetings at once. Products that send a bot into the call are the ones that need a back-office switch saying “no simultaneous recordings”. We don’t — not joining the call is the gate.
What expires is only the cloud copy of the original audio. Text evidence — bilingual quotes, timecodes and the bilingual minutes — is kept forever on every plan; the evidence links still open, there is simply no audio left to play. The original audio on your own machine is not governed by the cloud audio retention period: on the desktop app it stays on your own drive under the on-device cache limit you set — when that cache fills, the oldest already-uploaded segments are rotated out and nothing else ever touches it; the mobile app and browser capture release each local segment as soon as it has uploaded. Two weeks before anything expires, the Monday digest lists every meeting expiring in the next 14 days in a single email — we never send one email per meeting.
Never. No storage quota we impose is ever a precondition for meeting capture — that is an assertion written into the architecture, not a reassurance. Knowledge-base space is a fence, and it only governs the files you upload: when it is full, only new uploads stop. Existing files stay readable, searchable, citable and exportable — we never delete your files — and meeting capture, transcription and bilingual minutes are entirely unaffected. A fence takes no processing pack: when it is full you clean up, or you move up a plan (moving up raises the fence, it does not buy an allowance). A second limit caps how much you can ingest within one billing period at three times your space; deleting files does not return ingested volume.
Support
Because resolution time is not a quantity we control, and promising it would be a lie. How long a ticket takes depends on whether it is a configuration issue, a bug of ours, or an upstream vendor outage — the only thing we control is how long before a human actually sees it and answers you. So we commit to a first-response target only, and no plan commits to a resolution time. By the same logic these are targets, not guarantees, and a missed target produces no service credit, refund or other remedy — the industry attaches credits to availability, not to response clocks, because a response-clock payout bends the support team’s incentive toward firing off a holding reply to stop the bleeding. We publish the P1 attainment rate, not an average: averages get diluted by P3 volume and hide exactly the P1 failures that matter. Support hours are UTC 01:00–10:00, Monday to Friday; tickets filed outside them start the clock at the next window, and the clock pauses while we wait on your reply. Per-plan targets and all conditions are on the pricing page and in the Terms.
Probably not — our support day starts when yours ends (UTC 01:00–10:00, Monday to Friday). If your team spans time zones that is not necessarily bad news: file a P1 during your working hours and, typically, the reply is waiting for you when your next working day begins, instead of eating the hours you needed for meetings. Platform-wide incidents are a different matter — those are 24/7 on-call plus a status page, not a per-ticket response clock.
Teams & permissions
Meetings are private to you by default. You can share minutes with workspace members per meeting, with role-based permissions. Org-memory search respects the same permissions — everyone can only find meetings they are allowed to see.
Yes. External sharing creates a read-only link with an expiry and access scope you control, revocable anytime; every decision in a shared minute still carries its evidence link.
No matching questions — try another keyword, or just write to support@mlog.ai
Still have a question?
Write to support@mlog.ai or file a ticket in-app under Settings → Help & feedback. How soon we reply depends on your plan: the first-response targets are on the pricing page, and Starter has no response-time target — we still read every message. Requests about your personal data are acknowledged within one business day on every plan.
For business inquiries, write to business@mlog.ai