Use case · Operations

Weekly reports your boss actually reads

Hedy keeps an activity ledger all week — reviews shipped, questions answered, tasks closed, hours saved — then compiles the report nobody had to write. Each line links to the underlying activity records.

Hedy is a private, self-hosted AI employee. For weekly reports, it keeps a running activity ledger all week from your Lark/Slack, tickets, and repos, then turns one query into a finished report where every line cites its exact source. Unlike cloud AI seats that only summarize a chat thread, Hedy runs on your own Docker/K8s infrastructure, retrieves under the asker's ACL, and passes an evidence gate that quotes sources verbatim.

Last updated: 20 July 2026

How Hedy runs it

01

Ledger, not memory

Every deliverable is logged as it happens with links and cost. The report is a query, not a reconstruction.

02

Compiled and phrased

A weekly rollup with the interesting deltas highlighted — in prose a human enjoys.

03

Filed and delivered

To your document space and the right group, on schedule.

04

Auditable down to items

Each line links to the underlying activity records.

Governed by design

  • Same append-only ledger the audit view reads — one source of truth
  • Saved-hours estimates are labeled as estimates
  • Report delivery is an explicit, configured target

In depth

Most AI report tools start from a blank chat on Friday and reconstruct the week from memory. Hedy works the other way. It maintains an append-only activity ledger through the week, logging events as they happen across your connected systems. Lark is a first-class source alongside Slack. When you ask for the weekly report, Hedy is not guessing from a scrollback window. It is querying a ledger it already wrote, so the report reflects the full week, not the last thread that happened to stay in context.

One query orchestrates the report. You ask once, and Hedy composes sections, groups work by team or project, and drafts the narrative. This is the one顶一队 idea in practice: instead of adding an AI seat to each person's chat, a single Hedy produces the output a reporting team used to assemble by hand. The report is a work product, not a meeting recap. It is built to be read by whoever asked, filtered to what that person is allowed to see.

Every line is traceable. Hedy's evidence gate requires the answer to quote its sources verbatim, so each claim in the report links back to the ticket, message, or commit it came from. If a fact has no source in the ledger, it does not enter the report. Retrieval is ACL-scoped by the asker, so a manager's report and an IC's report drawn from the same ledger can legitimately differ in what they surface. Nothing leaks past the permission boundary just because it sits in the ledger.

Because Hedy is self-hosted, none of this leaves your infrastructure. You bring your own model key, including local vLLM or Ollama, so report content is never sent to a third party. There are no sub-processors. Deployment is a roughly 30-minute Docker Compose install, licensing is offline via Ed25519, and the ledger itself is append-only, which means the audit trail behind any weekly report cannot be silently rewritten after the fact.

Questions

How does Hedy make every line in a weekly report traceable?
Hedy runs behind an evidence gate: the report must quote its sources verbatim, and each claim links back to the originating ticket, message, or commit. Facts with no source in the activity ledger are excluded rather than invented. Because the ledger is append-only, the evidence behind any line cannot be silently altered after the report is generated.
Where does the weekly report data come from, and does it leave my infrastructure?
Hedy builds a running activity ledger through the week from your connected systems, with Lark/Feishu as a first-class source alongside Slack. Everything runs self-hosted on your own Docker or K8s, using your own model key including local vLLM or Ollama. There are no sub-processors, so report content never leaves your environment.
Can two people generate reports from the same ledger and see different things?
Yes. Retrieval is ACL-scoped by the person asking, so Hedy filters ledger content to what that individual is permitted to see. A manager's report and an IC's report drawn from the same week can legitimately differ. This keeps sensitive items from leaking across permission boundaries just because they exist in the shared ledger.
How is this different from a cloud AI that summarizes my chat?
Cloud assistants reconstruct the week from a chat scrollback at report time and live inside Slack or Teams. Hedy instead logs events into an append-only ledger all week, then turns one query into a full report with cited lines. It is a self-hosted AI employee producing a work product, not an assistant recapping a single thread.

Related

hedy.one

See it on your own infrastructure.

Book a demo