Use case · Operations

Walk into every meeting already briefed

Sixty seconds before the meeting, a pre-read lands in your direct messages: where the topic stands, what is at stake, what to ask, what to watch for. Read-only preparation — participants are never contacted.

Hedy's meeting pre-read is a self-hosted AI employee that builds a 60-second brief before any meeting: each agenda item's current status, what's at stake, your talking points, and risk flags. It runs read-only and never contacts attendees. Unlike cloud AI that lives in Slack and just records, Hedy is one AI employee doing a full analyst's prep on your own infrastructure, retrieving only what the asker is allowed to see.

Last updated: 20 July 2026

How Hedy runs it

01

Name the meeting

"Prep me for the 3pm with the platform team on the migration" — that is the whole prompt.

02

Context assembly

Issue status, related decisions, past meeting notes and org context — pulled and condensed.

03

Fifteen bullets max

Status, stakes, suggested talking points, risk flags. Scannable in one minute, hard-capped.

04

Linked evidence

Every bullet traces to its source. Trust, but verifiable.

Governed by design

  • Read-only preparation: participants are never contacted
  • Personnel notes inform tone but are never quoted into the brief
  • Briefs deliver to the requester only

In depth

A Hedy pre-read is not a transcript and not a note-taker. Before a meeting starts, one Hedy assembles a 60-second brief from the source material it is allowed to read: for each agenda item it states the current status, what is at stake, the talking points you should own, and any risk flags worth raising. It is the prep an analyst would do the night before, delivered as a single scannable page in Feishu/Lark or Slack. One Hedy does the work of a prep team, which is the whole point of the name: the one AI employee that replaces the headcount.

The pre-read is strictly read-only. Hedy does not message participants, does not reschedule, does not post to the meeting channel, and does not act on anyone's behalf. Reaching out to attendees is a write action, and every write action in Hedy is deny-by-default behind an explicit allowlist. So a briefing job can read calendars, docs, threads, and CRM records, but cannot send a single notification unless that specific action is whitelisted and approved. Governance is the product, not a setting you hope holds.

Retrieval is scoped by the asker. When you request a brief, Hedy searches under your ACL, so the pre-read never surfaces a document or thread you were not already entitled to read. The evidence gate requires every claim in the brief to quote its source verbatim, so a risk flag or a status line is traceable back to the exact record it came from, not a model guess. Every read and every retrieval lands in an append-only audit log you can inspect later.

It runs on your infrastructure. Hedy deploys to your own Docker or Kubernetes, including air-gapped, in about 30 minutes with compose. You bring your own model keys, including local vLLM or Ollama, so meeting content and source documents never leave your environment and there are no sub-processors. Licensing is offline via Ed25519, and pricing is per seat. Feishu/Lark is a first-class surface alongside Slack, so the brief lands where your team already works.

Questions

Will the AI message meeting participants or set up the meeting?
No. The pre-read is read-only by design. Contacting attendees, rescheduling, or posting to a channel are write actions, and in Hedy every write is deny-by-default behind an explicit allowlist. A briefing job can read your calendar, docs, and threads, but cannot send anything unless that action is whitelisted and approved.
Can the brief show me information I'm not supposed to see?
No. Retrieval is scoped to the person asking. Hedy searches under your ACL, so the pre-read only draws from documents and threads you are already entitled to read. The evidence gate also requires every line in the brief to quote its source verbatim, so each status and risk flag traces back to a specific record.
How is this different from a cloud meeting assistant that records calls?
Cloud assistants live in Slack or Teams and mostly capture what already happened. Hedy is a self-hosted AI employee that prepares before the meeting: agenda status, stakes, talking points, and risk flags in a 60-second read. It runs on your own infrastructure with your own model keys, and one Hedy does a full prep team's work rather than adding another seat that transcribes.
Where does the brief get delivered, and where does the data go?
The pre-read lands in Feishu/Lark or Slack, both first-class surfaces. All processing runs on your own Docker or Kubernetes, including air-gapped, deployable in about 30 minutes. You supply your own model keys, including local vLLM or Ollama, so source documents never leave your environment. There are no sub-processors, and licensing is offline via Ed25519.

Related

hedy.one

See it on your own infrastructure.

Book a demo