Coze gives you a studio to build bots and workflows. Hedy hands you a finished employee. The difference is whether you want to construct the worker — or hire one.
Hedy and Coze solve different problems. Coze is a platform where you build bots and workflows yourself—you design the flows, wire the nodes, and own every edge case. Hedy is a finished AI employee you hire: it runs self-hosted on your own infrastructure (Docker/K8s/air-gapped), speaks Lark and Slack as a first-class member, and ships governance built in—append-only audit, L0/L1/L2 approval tiers, deny-by-default writes, and per-asker ACL retrieval. With Coze you assemble capability and governance yourself. With Hedy, one hire replaces a team's output.
| Hedy | Coze | |
|---|---|---|
| What you get | A working AI employee, day one | A studio to build bots yourself |
| Autonomy model | Owns jobs: scheduled work, follow-ups, initiative | Responds to triggers and flows you design |
| Governance | Approval tiers, audit ledger, allowlists built-in | You assemble your own controls |
| Self-hosting | First-class: Compose / Helm / air-gap | Open-source studio version available |
| Knowledge | ACL-scoped RAG with mandatory citations | RAG blocks you configure per bot |
| Best at | Being a colleague in Feishu/Lark/Slack | Custom conversational apps and flows |
Based on public information as of July 2026. Corrections welcome: hello@hedy.one
Coze is a builder platform. You get nodes, prompts, plugins, and a canvas, and you assemble a bot or workflow. That is powerful when you want to construct something custom—but the burden of design, integration, and every failure mode stays with you. Hedy is the opposite starting point: it is a finished employee. You do not draw a flowchart; you assign work. It comes with a persona, skills, and scheduled duties, and it produces output the way a hired teammate would—closing the loop instead of handing you a toolkit.
The deeper split is governance. On a build-your-own platform, controls are something you remember to add: a check here, a guardrail there, mostly living in prompts. Hedy treats governance as the product. Write actions are deny-by-default with explicit allowlists. Approvals run in tiers—L0/L1/L2—so routine reads flow freely while consequential actions wait for a human. Retrieval is filtered by the asker's ACL, so people only get answers from documents they are entitled to see. An evidence gate requires every answer to quote its sources verbatim. And the audit log is append-only—no update or delete path. None of this is bolted on; it is how the runtime behaves.
Deployment and data boundaries differ too. Hedy is private, not cloud. It installs on your own infrastructure in about 30 minutes via Docker Compose, runs in air-gapped environments, and has no sub-processors—your data and prompts never leave your perimeter. You bring your own model keys, including local vLLM or Ollama, so inference can stay fully on-prem. Licensing is offline via Ed25519 signatures. A hosted bot platform routes your traffic and content through its cloud; Hedy is designed so nothing has to.
Fit is simple. Choose Coze when your goal is to build and operate bots yourself and you want a flexible construction surface. Choose Hedy when you want to hire an AI employee that shows up governed on day one—Lark and Slack native, self-hosted, audit-clean—and have one hire cover the output of a whole team rather than adding seats to a chat channel.