Meetnix

Roadmap

What is built, what is next, and what never will be.

Published because a self-hosted product asks you to commit to a direction, not a subscription. If a row here matters to you, say so — the order of the middle two sections is genuinely decided by who asks.

Everything the rest of this website describes in the present tense is in the Shipped section below. If you find a claim on any other page that is not, that is a bug on this site and we would like to know about it — hello@meetnix.ai. A roadmap page that quietly contradicts the front page is worse than no roadmap page. This one is the reference.

Shipped

In the build you can run today.

Shipped Per-track recording One file per participant per source, no compositing and no re-encode. The decision everything else rests on.
Shipped Identity-bearing manifests Principal, name and email per track, taken from the token your issuer signed, with a SHA-256 per file and a signature over the document.
Shipped Per-speaker transcription On hardware you nominate, with no diarisation step because nothing was ever mixed. One worker can serve several deployments.
Shipped Live captions Streaming transcription in the call, per speaker, from the same worker that writes the archive transcript.
Shipped Cross-meeting search Full-text over every transcript line, filtered by speaker, room or date, with permissions applied in the query rather than to the page.
Shipped OIDC sign-in and signed invites No second user database, no meeting passwords, and guest links that expire and are scoped to one room.
Shipped Retention policy and free-space floor Prune video early, keep audio and the transcript; disk pressure shortens video history rather than failing a recording.
Shipped Access audit log Append-only, covering playback, download, export, search and policy changes.
Shipped NixOS module, Docker image, Helm chart Three shapes, same binaries, same archive format. The NixOS path is the one we run and is tested by booting a machine.
Shipped Optional summaries Off by default. When on, only the text leaves, to a provider you configure — including one on your own hardware.
Shipped Operator branding Your name on the tab, the sign-in card and the mail, by config rather than by fork, and not behind a paid tier.

Building now

In the tree, not yet in a release.

Roughly the next two releases. These are being written; the dates we would give you would be wrong, so there are none.

Building Ask, with citations Plain-language questions over the archive where every sentence in the answer links to the track and offset it came from. Useless without the citation half, which is the part taking the time.
Building Sealed rooms End-to-end encrypted meetings with recording structurally impossible. The cryptography is the easy part; making the room's mode unmistakable to everyone who joins is not.
Building SIP dial-in and dial-out A phone caller with their own track and their own transcript lines, rather than a second-class participant in the record.
Building Live translation Second-language captions during the call, on your own hardware. Per-speaker input is what makes it usable rather than a demo.
Building Clips and highlights Cut a span of one speaker's track and export it with a manifest of its own, so where a clip came from travels with it.
Building Multi-tenancy and delegated administration One deployment, several organisations, storage and identity kept apart. What a group IT function needs before it can consolidate.

Next

Decided, scheduled, not started.

Next Native mobile apps iOS and Android. The browser works and is the right default; a scheduled call from a car is not a browser use case.
Next Multi-region SFU with failover Schedule media near the participants and the recorder near the data. They do not have to be the same place, and for some buyers they legally cannot be.
Next Transcript redaction Remove a passage from the transcript and the corresponding audio, leaving a recorded gap rather than a silent edit. Erasure that is visible is the only kind worth having.
Next Webhooks and an API Meeting-ended events, transcript-ready events, and a read API over the archive, so the record can feed the systems that should have it.
Next Chat and ticketing integrations Slack, Teams, Matrix and the issue tracker: a link to the meeting, the summary, and the two lines where a decision was made.
Next Recording of the room, not just the call A hardware appliance for the conference room, so the six people around one table are six tracks rather than one shared microphone.

Considering

Argued about internally, not committed to.

Maybe Federation between instances Two companies, two Meetnix deployments, one meeting, and each side's recording staying on its own side. Technically clear, politically interesting, and the manifest would have to say which half it holds.
Maybe On-device transcription Transcribe in the browser so the audio never reaches the server either. Would weaken the archive; would strengthen the sealed room.
Maybe Speaker verification Confirming the voice on a track matches the account that signed in. Genuinely useful in a dispute, and one careless step from being the biometric surveillance we say we are against.
Maybe Asynchronous video Recorded updates with the same per-track archive and the same search. A different product wearing this one's clothes, possibly a better one.

Never

Decided against, permanently.

A roadmap is more useful for what it rules out. These are not missing features; they are the ones a competitor will ship and we will not, and you should hold us to that.

Attention and engagement scoring

No "was this person looking at the window", no talk-time leaderboard framed as coaching. The record is for what was said, not for grading who said it.

Sentiment and emotion inference

Reading a mood off a face or a voice is unreliable, unfalsifiable and lands hardest on the people with the least power in the meeting.

Silent or invisible recording

No hidden mode, no admin override, no flag that records without every client showing it. If we ever want this, the honest move is to say so loudly, not to add a checkbox.

Telemetry or a licence check

A self-hosted instance makes no call to us, ever. Not for usage counts, not for updates, not to verify a seat count. Air-gapped has to keep working.

Training on your archive

Not on Cloud either. The media is in your bucket and we have no business reason to read it — a promise that is worth more when the architecture makes it awkward to break.

Crippling the free build

No user cap, no watermark, no meeting length limit, no feature removed to make a demo call go badly. If we cannot sell the terms and the support, we do not deserve the revenue.

Move something up the list.

The middle two sections are ordered by who asks. Tell us which row you need and roughly when, and we will tell you honestly whether that is realistic.