Use case · Deployment

The AI employee that speaks Feishu and Lark natively

Most AI employees stop at Slack. Hedy is first-class in Feishu and Lark: direct and group chat, approval cards, cloud-doc filing, org-directory awareness — plus Slack when you need it. Both CN and international endpoints are supported.

Hedy is a self-hosted AI employee that lives natively inside Feishu and Lark, not just Slack. It works in DMs and group chats, posts approval cards for L0/L1/L2 sign-off, archives outputs to Feishu Docs, and reads the org directory to enforce per-asker ACL retrieval. It runs on both the China Feishu tenant and international Lark. Most cloud rivals only integrate Slack; Hedy covers both ends and stays on your infrastructure.

Last updated: 20 July 2026

How Hedy runs it

01

Real membership

Hedy joins allowlisted groups with her own name and account, answering when mentioned.

02

Approval cards in chat

Outward actions arrive as one-tap approve or deny cards — governance without leaving the thread.

03

Cloud-doc filing

Minutes, reports and research file straight into Feishu or Lark docs with links back.

04

Org-aware

Directory sync means Hedy knows departments and reporting lines when routing or drafting.

Governed by design

  • Both Feishu (CN) and Lark (international) endpoints are supported
  • Direct-message and group access are allowlists — she talks only where placed
  • Every message she sends is logged and attributable

In depth

Hedy is a full first-class citizen inside Feishu and Lark, not a bot bolted onto Slack. It answers in direct messages, joins group chats, and delivers work through native approval cards. When Hedy proposes a write action, it renders an interactive card so an approver taps to grant or deny, and every decision routes through the L0/L1/L2 approval tiers. Slack is supported too, but Feishu and Lark are where Hedy does its primary work, which matters if your company already runs its day on Lark.

Directory awareness is what makes this safe at scale. Hedy reads the Feishu/Lark contacts graph, so retrieval is filtered per asker: a person only ever sees answers grounded in sources their own ACL permits. Nobody escalates their access by asking Hedy a clever question. Finished outputs are archived to Feishu Docs automatically, so the work lands where your team already reads and reviews, with an append-only audit trail behind it.

The dual-tenant reality is the differentiator. Feishu (国内) and Lark (international) are separate ecosystems, and most AI-employee products integrate neither, only Slack. Hedy speaks both from one deployment, so a company operating across mainland China and overseas offices runs a single AI employee instead of stitching together tools per region. This is the one in hedy.one: one AI employee covering the whole surface, not a per-seat swarm.

Because Hedy is self-hosted, all of this stays on your infrastructure via Docker or K8s, deployable in about 30 minutes with compose. You bring your own model keys, including local vLLM or Ollama, so no chat content or documents leave your network and there are no sub-processors. Governance is the product, not a setting: write operations default to deny with an explicit allowlist, red-line behaviors are tested, and the evidence gate forces every answer to quote its sources verbatim.

Questions

Does Hedy work with Feishu and Lark, or only Slack?
Both Feishu (国内) and international Lark are first-class citizens: DMs, group chats, native approval cards, and Feishu Docs archiving all work out of the box. Slack is also supported. Many competing AI-employee products integrate Slack only, so Hedy covers the two Feishu/Lark tenants that most rivals do not.
How does Hedy handle permissions inside a Feishu group chat?
Hedy is directory-aware. It reads the Feishu/Lark contacts graph and filters retrieval per asker, so each person only receives answers grounded in sources their own ACL allows. Write actions default to deny and require an explicit allowlist plus an L0/L1/L2 approval card, so no one gains access by asking in a shared channel.
Where do Hedy's outputs go after it finishes a task?
Completed work is archived to Feishu Docs automatically, landing where your team already reviews. Every action is recorded in an append-only audit log, and the evidence gate requires each answer to quote its source material verbatim, so archived outputs stay traceable to their origin.
Can Hedy run entirely on our own infrastructure?
Yes. Hedy is self-hosted via Docker or Kubernetes, deployable in about 30 minutes with compose, including air-gapped setups. You supply your own model keys, including local vLLM or Ollama, so chat and document data never leave your network. There are no sub-processors, and licensing is offline via Ed25519.

Related

hedy.one

See it on your own infrastructure.

Book a demo