The short answer: office AI agents are not one product category. Platform-native tools (DingTalk, Feishu, WeCom) sell existing collaboration data plus AI. Standalone agents (WorkBuddy) sell task execution across applications. Custom builds (LingWork) sell depth inside a defined business boundary. Do not compare annual fees or seat prices first. Compare data context, execution permissions, AI quota, and exit cost.
This is a comparison and procurement analysis for companies with 50–1,000 people, CIOs and IT teams, and admin or operations leads. Research cut-off is 25 October 2026. Prices are official public list prices and purchase pages, in RMB, shown per year or per seat. They are not a formal quote from Lingyi or any vendor. Tax, discounts, implementation, training and overage depend on the contract in force.
Task first, then platform. Incremental cost first, then total price. Exit first, then signature. Platform-native, standalone agent and custom build are not the same quotation.
What kinds of office AI agents are there?
There are three paths: platform-native, standalone agent, and custom build. The first is strongest on collaboration data, the second on cross-application execution, and the third on business fit and boundaries. Start by deciding where the task belongs.
| Path | Examples | Strengths | Main boundary | Best fit |
|---|---|---|---|---|
| Platform-native | DingTalk, Feishu, WeCom | Ready IM, OA, document and meeting context; mature permissions and org structure | Limited cross-platform reach; AI editions, quotas and seats are priced in complex ways | Tasks mostly happen inside the platform; fast rollout |
| Standalone agent | WorkBuddy and similar | Cross-application execution; desktop, document and browser tasks; seat plus Credits | Accounts and permissions must be connected; data boundary depends on deployment | Knowledge work with frequent cross-system retrieval, generation and cleanup |
| Custom build | LingWork, in-house builds | Designed around business processes; permissions, sandbox and interfaces can be specified | High first-year development and integration cost; ongoing operations | Core systems, private data, deep processes or multi-system coordination |
The categories are editorial, not a vendor taxonomy. One product can have both platform-native and standalone capabilities.
Where do public prices for DingTalk, Feishu, WeCom and WorkBuddy differ?
Feishu AI is RMB 9,900 or RMB 99,000 per year. WorkBuddy is RMB 198 or RMB 316 per seat per month. DingTalk AI editions are mostly quote-only. WeCom publishes no unified AI price list. The units differ, so normalize them before comparing.
| Product | Public price | Included or metered | Extra or unknown |
|---|---|---|---|
| Feishu AI | Basic RMB 9,900/year; Enterprise RMB 99,000/year | Basic includes 180,000 AI points; Enterprise includes 2 million points | Feishu Business or Enterprise suite is separate; Plus and Flagship follow the official page; overage is separate |
| DingTalk AI | AI Basic, Advanced, Premium and Flagship: contact sales. AI Assistant: Professional edition plus AI Assistant at RMB 19,800 per organization per year; Dedicated edition plus AI Assistant from RMB 118,000 per organization per year | Editions include different AI quotas, storage, meetings and APIs | Seats, editions, add-ons and implementation are separate; AI commercial editions changed from 21 October 2026 |
| WeCom AI | No unified AI edition price on the official public page | Smart assistant, smart bot, smart search, smart summary and similar features | AI entitlements, conversation archiving, third-party services and partner quotes need confirmation |
| WorkBuddy Enterprise | Flagship RMB 198/seat/month (RMB 2,376/seat/year); Dedicated RMB 316/seat/month (RMB 3,792/seat/year, 100 seats minimum) | Seat subscription plus Credits; admin console, permissions and audit | Overage Credits, private deployment, implementation and training are separate |
| LingWork / custom | Project-based, no list price | Scoped to scenario, interfaces, deployment and permissions | Development, integration, models and tools, and operations are separate |
Prices were accessed on 25 October 2026. Feishu, DingTalk and WorkBuddy prices come from public pages. WeCom is listed as having no unified public AI price, which does not mean it is free. All prices can change; confirm on the official page or in the contract. AI quota units differ, and Feishu points cannot be converted directly into WorkBuddy Credits.
Why can't you compare annual fees and seat prices directly?
Suite inclusion, AI quotas, seat versus organization pricing, overage, implementation and permission scope all differ. At the same headcount of 100, annual fees can differ several-fold, and so can the tasks covered.
| Dimension | Platform-native | Standalone agent | Custom build |
|---|---|---|---|
| Suite inclusion | Usually layered on an existing suite, which is priced separately | Seat subscription; some office suites include it | Project-based, may include development and deployment |
| AI quota | Points per year; overage purchased separately | Credits per seat per month; overage purchased separately | Actual model, call and resource cost |
| Pricing unit | A mix of organization, edition and seat | Seat plus Credits | Project plus operations |
| Execution permissions | Mostly apps inside the platform | Cross-application, desktop and browser | Defined by system interfaces |
| Implementation and exit | Fast rollout; migration limited by the platform | Accounts and permissions must be connected; exit depends on data export | High first-year investment; exit and migration need contract terms |
This table explains pricing differences; it is not a ranking of products. Editions from the same vendor also change the boundary.
What does a 100-person company pay per year?
At public list prices, Feishu AI Basic plus the Standard suite is about RMB 69,900 a year, and Feishu AI Enterprise about RMB 159,000. WorkBuddy at 100 seats is about RMB 237,600 a year. DingTalk and WeCom require a quote and cannot be modeled by formula. Implementation and overage are excluded in every case.
| Option | First-year calculation | Result | Notes |
|---|---|---|---|
| Feishu AI Basic plus Standard suite | 9,900 + 50 × 100 × 12 | RMB 69,900/year | Standard suite at RMB 50/person/month; if the suite is already in use, the increment is only RMB 9,900 |
| Feishu AI Enterprise plus Standard suite | 99,000 + 50 × 100 × 12 | RMB 159,000/year | Includes 2 million AI points; overage is separate |
| WorkBuddy Flagship | 198 × 100 × 12 | RMB 237,600/year | 100 seats; Credits beyond the allowance are purchased separately |
| WorkBuddy Dedicated | 316 × 100 × 12 | RMB 379,200/year | 100-seat minimum; VPC or dedicated deployment, with terms per contract |
| DingTalk / WeCom / LingWork | By quote | Cannot be modeled uniformly | Editions, seats, quotas and implementation vary widely |
The examples assume RMB, 100 people, 12 months and public list prices, excluding tax, discounts, implementation, training, integration and overage. The Feishu suite uses the Standard edition at RMB 50/person/month as an illustration; the actual edition and discount follow the purchase page. These are not quotes.
Platform-native, standalone agent or custom build?
If you already run a platform and the task lives in IM or OA, choose native AI. If work crosses applications and is knowledge-heavy, choose a standalone agent. If compliance is strict, processes are deep or many systems are involved, choose a custom build. Choose by task boundary, not by brand.
| Scenario | Recommendation | Why | Avoid |
|---|---|---|---|
| Already deep in Feishu, DingTalk or WeCom; tasks sit in approvals, documents and meetings | Platform-native AI | Context and org permissions are ready; fast rollout | Buying a separate agent for duplicate Q&A |
| Retrieval and cleanup across documents, email, calendar, cloud drive and CRM | Standalone agent | Stronger cross-application execution and desktop actions | Platform-only Q&A that leaves people moving data by hand |
| Write-back to core systems, private data, strong audit, multiple systems | Custom build / LingWork | Permissions, sandbox, interfaces and processes can be tailored | Buying standard SaaS to carry core write-back |
| Customer service or private-domain operations already in WeCom | WeCom AI plus a standalone agent | Customer context lives in WeCom; complex tasks can go to the agent | Copying customer data into external tools |
| Under 50 people, single Q&A or summary | Existing platform AI, free or basic | The task is simple; a standalone agent adds limited value | Building in-house or buying seats for everyone at the start |
The selection matrix is editorial advice, not a vendor ranking. Validate it against tasks, data, permissions and budget.
What to ask at purchase, and how do you avoid lock-in?
Ask whether the suite is included, how AI quotas and overage work, whether pricing is per seat or per organization, what data, model and tool permissions apply, whether private deployment is possible, how logging and audit work, how data is exported on exit, and what the SLA covers. Put the answers in the contract.
| Question | Why ask | Red flag |
|---|---|---|
| Does the AI edition include the suite? | Avoid seeing only the AI fee and missing the suite or seats | A quote that says "AI 9,900" without suite conditions |
| How are AI quotas metered, and what does overage cost? | Points and Credits use different units; overage can be expensive | Only a total quota, with no per-task consumption |
| Is pricing per seat or per organization? | A 100-person plan can differ several-fold | Mixing organization and seat prices |
| Which data and tools can it access? | Broader permissions raise leak and privilege-escalation risk | Full permissions by default, with no least-privilege design |
| Can it run privately or hybrid? | Essential when core data cannot leave the boundary | "Private" in name, but key models or tools still call the public internet |
| Logs, audit and human takeover? | Incidents must be traceable, blockable and reviewable | Only aggregate views, with no per-task record |
| Can you export on exit? | Avoid locked data, configuration and history | No export of conversations, knowledge base or configuration |
The question list is procurement advice, not an assessment of any vendor. Business, legal and security should review the terms together.
Which companies should not buy yet?
Companies without a defined task, without a data or content owner, not using the AI already in their platform, without permissions and audit ready, or buying only the "agent" label should wait.
- Unclear task: you cannot name a weekly, repetitive, time-consuming task that can be accepted. The tool becomes a demo.
- No owner: nobody owns the knowledge base, permissions, evaluation and operations. The tool is soon idle.
- Existing AI unused: the platform's Q&A, summaries and meeting notes are not in use. Improve those first.
- Permissions not ready: you do not know which data the agent can read or which systems it can write to. Do not connect core business yet.
- Label-driven: treating "do we have an agent" as a KPI instead of task closure and cost.
Who is LingWork for, and who is it not for?
It fits companies with cross-application tasks that need sandboxing, permissions and custom processes. It does not fit teams that only need Q&A, have no IT or content owner, or compare only seat prices.
The LingWork office agent product page emphasizes one-sentence tasks, agent planning and execution, office and personal modes, a secure sandbox, the MCP tool ecosystem and document generation. It is built for handing multi-application, multi-step office work to an agent, with clear requirements for permissions and boundaries.
Not suitable for: teams that only need meeting notes or document Q&A, have no content owner, have budget for a single IM seat, or expect a zero-operations product out of the box. LingWork does not publish a list price on this page because price depends on scenario, interfaces, deployment and permission design. Confirm through an assessment and contract.
If you are evaluating agent permissions, classification and procurement compliance, see China's 2026 agent rules: a practical compliance checklist. For post-launch operating metrics, see AI agent metrics in 2026: what production data really shows. For the infrastructure budget behind private deployment, see What private AI deployment really costs in 2026.
FAQ
Does Feishu AI at RMB 9,900 per year include the Feishu suite?
The public pages present Feishu AI editions separately from the Business and Enterprise suites. The RMB 9,900 AI Basic package includes 180,000 AI points; the suite is billed separately. Whether your existing suite qualifies and how overage is charged must be confirmed on the current Feishu purchase page and admin console.
How much does a DingTalk AI edition cost?
DingTalk's public pages for AI Basic, Advanced, Premium and Flagship mostly say "contact sales", with no unified published annual fee. The AI Assistant page separately lists the Professional edition plus AI Assistant at RMB 19,800 per organization per year, and the Dedicated edition plus AI Assistant from RMB 118,000 per organization per year. The two scopes differ, so confirm the edition, seats and entitlements before buying.
Is WeCom AI free?
WeCom's official pages offer a smart assistant, smart bot, smart search, smart summary and other AI features, but publish no unified AI edition price. Basic registration, specific AI entitlements, conversation archiving and third-party services may be charged separately. Confirm with the WeCom admin console, partner quote and contract. Do not assume "all free".
Is WorkBuddy at RMB 198 per seat per month expensive?
Do not look at the seat price alone. RMB 198 per seat per month is the Enterprise Flagship edition, and RMB 316 is Enterprise Dedicated (100 seats minimum). Both involve Credits and overage purchases. Compare cost per completed task against the work it replaces, the labor it saves and the systems it connects, not against an IM suite's annual fee.
Does a 100-person company need 100 office agent seats?
No. Start with 10–30 frequent users or one or two scenarios. Expand based on weekly active use, task completion rate and cost per task. Buying for everyone at once creates idle seats and increases permission and data risk.
Is a custom office agent always more expensive than SaaS?
Not necessarily, but the cost structure differs. Custom work has high first-year development, integration and security costs, then ongoing operations and model or tool fees. SaaS is usually faster for standard office tasks. Custom or tailored work makes more sense for core systems, private data and deep processes. Compare three-year TCO and task closure.
Choosing an office agent is not "DingTalk or Feishu". It is "where does the task live, where does the data live, how far do permissions go, and how do we exit". Answer those four before discussing annual fees and seats.
Sources and caveats
Research cut-off: 25 October 2026. This article is a review of public material and editorial modeling, not first-hand procurement, a trial or vendor certification. Prices come from the public pages of Feishu, DingTalk, WeCom and Tencent Cloud WorkBuddy, all accessed on 25 October 2026. Pages can change; discounts, tax, implementation, training, overage and contract terms alter the actual cost. WeCom publishes no unified AI price, and we did not estimate one for it. The 100-person examples are editorial assumptions, not quotes. LingWork has no list price on this page; product capability follows the product page and contract. We do not promise savings, rankings or procurement outcomes.
- Feishu, "Get Feishu AI now" and "Feishu pricing," accessed 25 October 2026. AI Basic RMB 9,900/year (180,000 points); Enterprise RMB 99,000/year (2 million points); Business and Enterprise suites from RMB 50/80/120 per person per month. AI and suites are shown separately. feishu.cn/service/ai, feishu.cn/price-list.
- DingTalk, "AI Assistant" and the DingTalk open platform and official store, accessed 25 October 2026. AI Assistant: Professional edition plus AI Assistant RMB 19,800 per organization per year; Dedicated edition plus AI Assistant from RMB 118,000 per organization per year. AI Basic, Advanced, Premium and Flagship pages are mostly quote-only. AI commercial editions changed from 21 October 2026. page.dingtalk.com.
- WeCom, "AI for smart office," accessed 25 October 2026. The official page presents smart assistant, smart bot, smart search and smart summary features, without a unified AI edition price. Entitlements and prices follow the WeCom admin console and service partners. work.weixin.qq.com.
- Tencent Cloud, "WorkBuddy Enterprise billing and purchase guide," updated 28 September 2026, PDF 23 October 2026. Enterprise Flagship RMB 198/person/month (RMB 2,376/person/year); Enterprise Dedicated RMB 316/person/month (RMB 3,792/person/year, 100 seats minimum); seat plus Credits. cloud.tencent.com.
- Lingyi Information, LingWork product page, accessed 25 October 2026. Product capability description, not a third-party review. LingWork is priced per project; this article does not publish a list price. ly-work.html.