DingTalk open-sources DWS, a command-line tool that lets AI agents run workplace tasks
Alibaba's DingTalk released its workspace command-line tool under Apache 2.0, with admin-gated access and a HarmonyOS PC build, so agents can act on office data.
The short version
- DWS is on GitHub under the Apache 2.0 license, built for both people and AI agents.
- Per IT Home, it spans 25 business modules and more than 1,000 commands.
- Output is structured JSON, with a dry-run preview and a field filter to save tokens.
- An enterprise administrator must authorize it; calls pass through DingTalk's authentication and audit chain.
- The HarmonyOS PC build lacks some features, including SafeChat and runtime native payloads.
Why it matters
Agents become useful when they can do things in the software you already use, and a command-line tool is one of the plainest ways to allow that. Because DWS is open source, a company can read what it does before giving an agent access to calendars, approvals and documents. The release also shows how permissions, audit trails and preview modes are becoming part of agent tooling.
What DingTalk released
As pandaily.com reported, DingTalk, Alibaba's workplace platform, has open-sourced DWS, short for DingTalk Workspace. It is a command-line tool: a program you operate by typing short text instructions instead of clicking through an app. The code is on GitHub under the Apache 2.0 license, which allows broad use and modification.
The project presents itself as DingTalk's workspace delivered through the command line, designed for human users and AI agents alike. An agent here means an AI program that carries out tasks on a user's behalf rather than just answering questions.
IT Home, cited by Pandaily, says DWS covers 25 business modules and more than 1,000 commands. Those cover most of what an office runs on: the address book, messaging and group admin, calendars and meeting rooms, task lists, approval flows, attendance logs, work reports, documents, whiteboards, file storage, AI-generated meeting notes and email.
Why a command line suits agents
The intended workflow is simple. Rather than opening several screens to move through messages and tasks, a user tells an agent the goal, and the agent works out which commands to run. It acts only within the permissions it has been given.
Several design choices are aimed at software readers. Commands return JSON, a tidy, predictable text format that programs parse easily. The tool also ships with built-in agent skills, which are packaged instructions that teach an agent how to use a product. You can install them for one product at a time or as a single combined bundle.
The documentation also suggests three habits for agent use. One is a flag that skips confirmation prompts, since an agent has no one to click "yes." Another is a dry-run mode that previews a request without carrying it out. The third is a filter that returns only the fields you need, so the agent reads less text and uses fewer tokens, the small units of text that AI models are billed and limited by.
Guardrails around company data
A tool that can read approvals and attendance records is sensitive, and the project treats it that way. For now, DWS needs authorization from an enterprise administrator, so an individual employee cannot simply switch it on.
The listed protections include:
- Login through OAuth device flow, a standard way to sign in from a program without typing a password into it.
- Encrypted storage of access tokens, tied to the specific device.
- A domain allowlist that, by default, sends credentials only to DingTalk's own domains.
- Required HTTPS connections.
- Filters against path traversal and Unicode spoofing, tricks that use crafted file paths or look-alike characters to fool a program. They are meant to keep an agent from being misled by malicious input.
DingTalk also says every call goes through the authentication and audit chain of its open platform. That means actions taken by an agent should leave a record that administrators can review.
The HarmonyOS PC build
The release also targets Huawei's desktop platform. The support comes through OpenHarmony, the open-source base beneath HarmonyOS. The repository can produce a binary for the arm64 chip architecture using an OpenHarmony version of the Go toolchain, the set of tools that compiles Go code.
According to the project, each beta and stable release now builds, signs, verifies and publishes this package, and archives for six other platforms ship with it. The project says the binary has been tested on a HarmonyOS PC.
It is not a full match for the other builds. SafeChat and runtime native payloads are unavailable there. IT Home calls this the first open-source office command-line tool on HarmonyOS PCs, and says it fills a gap in productivity software for that desktop system.
What to keep in mind
The details above come from Pandaily's report and the project's own documentation, as relayed by IT Home. They describe what the tool is designed to do, not how it behaves in a real company. We have not independently tested it.
The admin-approval requirement is both a strength and a limit. It keeps control with the organization, but it also means the tool is aimed at teams already using DingTalk, not at anyone who wants to experiment on their own.
For anyone building agents, the pattern is worth noticing: open code that can be inspected, a way to preview actions, narrow permissions, and an audit trail. Those are the same questions to ask of any tool you hand to an agent.
What to watch
- Will DingTalk loosen the administrator-only authorization, and will the HarmonyOS build gain the missing features?
- Do other workplace platforms release command-line tools built for agents, and with what safeguards?
- How do the audit logs and permission limits hold up once real companies run agents against live data?
Try it
If you plan to give an agent access to any work tool, check three things first: whether it has a dry-run mode, whether permissions can be limited, and whether actions are logged.
Sources and links
- Reportingpandaily.com: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by pandaily.com · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
