Products · Anchor · Beta
Midcore Meet.
Communication and meetings, AI-native by construction.
A standalone communication platform — threads, calls, meetings, directory — where capture, memory, and context sit inside your trust domain instead of a vendor’s. Incumbents bolt AI onto a product whose primary objects are rooms and messages. Meet was built the other way round.
Why it is different
AI-native, not AI-added.
The incumbents were built when the primary objects of a communication product were rooms and messages. Everything since — transcription, summaries, assistants — has been fitted on top of that shape. It works, and it leaves the transcript and the model in a vendor's trust domain rather than yours.
Meet starts from different objects. Memory, context, and authority are first-class — a meeting produces episodic facts with provenance, context assembly cites its sources, and every consequential action passes an approval surface that leaves a receipt. Capture and reasoning sit in one system, which is what makes the audit trail verifiable rather than asserted.
It is younger and smaller than what it hopes to replace. The section below is scoped to what runs today.
What ships today
Running code, not a roadmap.
Every line below maps to a surface you can open in a live walkthrough. Anything still being built is in the roadmap further down — never implied here by omission.
Conversation
- Threads and direct messages with presence, typing, and read receipts
- Forward, favourites, reply threads, and full-text search
- Attachments with preview, and a command palette over everything
Meetings and calls
- Video calls with screen share, breakout rooms, and a waiting room
- Host controls, participant roster, and ring/knock handling
- Scheduling with calendar view and ICS import
- Browser guest join — external participants install nothing
Memory and context
- Meeting transcription with chain-of-custody and encrypted spill
- Episodic memory with provenance and retention policy
- Context assembly that cites the source of every claim
- Company directory with org chart, reporting chain, and peers
Governance
- Approval workflows with capability delegation and grant templates
- Durable multi-day tasks with handoff and outcome digests
- Tamper-evident hash-chained audit with independent verification
- Subject erasure that cascades through derived memory
The trust surface
Consent is an affordance, not a setting.
An AI that remembers a conversation permanently poses a different question from a recording. Meet answers it in the interface, at the moment it matters.
Disclosure at the point of presence
An AI participant is announced on its own tile and in the call trust rail — not buried in an admin policy nobody in the room has read.
Per-meeting memory opt-out
Any participant can take a single meeting out of retention, without leaving the call and without an administrator.
Audit you can verify yourself
Consequential events are hash-chained, and the chain has an independent verification path. An auditor does not have to take our logs on faith.
Erasure that actually propagates
Deleting a subject cascades into derived memory, rather than leaving the facts an assistant learned sitting behind a tombstone.
On the roadmap
Not available yet. Stated plainly.
These are in active development and are not part of the beta. We list them so the boundary is unambiguous — dates are targets, not commitments.
- Q3 2026
AI colleagues as meeting attendees
Agents are already first-class in the directory and appear in the org chart. Inviting one to a meeting as a named, disclosed attendee is the next step.
- Q3 2026
Propagation across agents
Correct a fact once — "Sara left that project in May" — and have the correction reach every agent that relied on it, rather than each holding its own stale copy.
- Q4 2026
Authoritative org model
Effective-dated positions, units, and reporting lines as a first-class schema, so the org chart is a source of truth rather than a derived view.
- Q4 2026
Scale substrate
A Postgres substrate with row-level tenant isolation, for deployments beyond the single-instance footprint Meet targets today.
FAQ
Questions, frankly answered.
- Is Midcore Meet available today?
- It is in beta. Everything in "What ships today" is running code you can be shown in a live call. Everything not yet built is in the roadmap section rather than implied by omission — we would rather lose a deal than win one on a feature that does not exist.
- Can an AI employee attend my meeting?
- Not yet, and we will not pretend otherwise. Agents are first-class in the directory and appear in your org chart today, but attending a scheduled meeting as a named participant is a Q3 2026 item. When it lands it will arrive with disclosure and a per-meeting memory opt-out, not quietly.
- Do participants know when an AI is present?
- Yes. An AI participant is disclosed on its tile and in the call trust rail, and any participant can opt that meeting out of memory retention. We treat an AI that remembers permanently as a different consent question from a recording, because it is one.
- Where does meeting data live?
- Inside your trust domain. Capture, transcript, memory, and context assembly are part of the same system rather than split between your tenant and a vendor’s model host. That is the architectural difference; the rest follows from it.
- How does this compare to Teams, Zoom, or Google Meet?
- Those are excellent human-centric products with AI added afterwards, and their primary objects are rooms and messages. Meet is smaller and younger. What it offers instead is that memory, context, and authority are first-class objects rather than features, and that the audit trail is verifiable rather than asserted.
Next step
See it running.
The fastest way to judge Meet is a live call inside Meet. Bring the awkward questions.