Global SaaS plus DingTalk and Feishu: one AI assistant
Companies operating across markets often run two stacks: overseas teams on global SaaS and email, teams in China on DingTalk, Feishu or WeCom. When adding an AI assistant, there is no need to build one for each side. Map who asks from which entry point and which systems they need; map both sets of accounts to one directory and one set of permissions; decide where the assistant runs, and which data stays local, based on where the data lives; answer in Chinese or English with sources shown; and have a person confirm before any data changes. Pilot one process that spans both sides, then expand.
Companies working across markets often have two sets of collaboration tools. Overseas teams use global CRM, ticketing, online documents and email; teams in China use DingTalk, Feishu or WeCom; finance and ERP may sit somewhere else again. When an AI assistant comes in, the easy path is to build one on each side. The result is two knowledge bases and two different answers to the same question.
A steadier approach is to run one assistant, connect both sides' collaboration tools as entry points, and connect both sides' business systems as tools it can call. Before rollout, settle five things: an entry-point and system map, identity and permissions, where data lives, language and sources, and confirmation before data changes.
1. Map entry points and systems first
List both sides in one table:
| What to list | Overseas teams | Teams in China |
|---|---|---|
| Where they ask | Global collaboration tools, email | DingTalk, Feishu or WeCom |
| Systems to query | Global CRM, ticketing, online documents | ERP, OA, finance systems |
| Actions needed | E.g. update a customer record, open a ticket | E.g. start an approval, look up an order |
| Main language | Mostly English | Mostly Chinese |
With the map in hand, you can see which questions belong to one side and which span both, such as an overseas salesperson asking about shipping status for an order handled in China. Questions that span both sides are exactly where one assistant beats two.
2. One person, one set of permissions
The same employee may have an account in a global collaboration tool and another on a Chinese platform. The assistant should treat them as one person with one set of permissions: whichever entry point they use, they can see the same scope, no more and no less.
Set permissions by person, not by entry point. Before launch, align both sets of accounts with the company directory, and sign off on that step separately. For connecting the three Chinese collaboration platforms at once, see One AI assistant for DingTalk, Feishu and WeCom.
3. Let data location decide where it runs
Companies working across borders face one extra question: which country or region the data sits in, and whether it may cross borders. Have legal or compliance confirm this first, then decide where the assistant runs, which systems' data is handled only locally, and what may be combined.
A common arrangement is to run the assistant in an environment the company controls and limit what it can call based on where the data lives. For data with cross-border restrictions, answer locally and do not pull it to the other side. Where the models the assistant calls are hosted also counts toward where data goes.
4. Answer in Chinese or English, with sources
Overseas colleagues ask in English and colleagues in China ask in Chinese, so the assistant should answer in the language of the question and show which system and which document each point comes from. When the two sides' material differs, for example English and Chinese versions of a policy updated at different times, the answer should show which version it relied on.
5. Have a person confirm before data changes
Queries and drafts can be looser. Once the assistant is about to change a record, start an approval or send an external message, it should hand the proposed action to a person, act only after confirmation, and keep a record. Write operations from either side follow the same confirmation rules. For how confirmation is designed, see Where human review sits.
Pilot one process that spans both sides
There is no need to connect everything on day one. Pick a fairly fixed process that spans both sides, such as an overseas salesperson asking for an order status in their collaboration tool and the assistant checking the ERP in China and replying in English. Once that works, add entry points and systems step by step.
FAQ
**Can overseas and China-based teams share one AI assistant?**Yes. Run one assistant, connect both sides' collaboration tools as entry points and both sides' business systems as tools behind it, and maintain knowledge and permissions in one place.
**How is cross-border data handled?**Legal or compliance first confirms which data may cross borders; the assistant's callable scope follows where the data lives, and restricted data is answered locally without being pulled to the other side.
**With two languages, how do answers stay consistent?**Answers come in the language of the question and cite the system and document used; where Chinese and English material differ, check against the source shown in the answer.
**What does FIM One by Beijing FIM Network Technology Co., Ltd. do?**FIM One is a self-hosted agent platform for companies that use both global SaaS and DingTalk, WeCom and Feishu: it connects existing systems, orchestrates tasks across them, pauses write operations for a person to confirm, and records every operation. FIM One is source-available software and works with any OpenAI-compatible model endpoint.
Summary
For a company working across markets, run one AI assistant for both sides: map entry points and systems, map accounts to one set of permissions, let data location decide where it runs and what it can call, answer in Chinese or English with sources, and have a person confirm before data changes. Pilot one process that spans both sides, then expand.
FIM One by Beijing FIM Network Technology Co., Ltd. is built for companies that use both global SaaS and Chinese collaboration platforms. It can be self-hosted, and the Community edition is complete and free forever. Learn more on About FIM One.