← → flip page
AIPaddle · Enterprise AI Harness

Create a Super AI Individual,
Organization formedSuper AI organization

Employees already use AI. The challenge is to retain those capabilities within the enterprise, make them reusable, coordinate them under clear rules, and govern, audit, and measure their impact. AIPaddle is the Enterprise AI Harness built for that purpose.

Capture · AI capabilities become organizational assets Collaboration · Human and AI employees coexist in the same role Governance · Authorized, auditable, measurable Native · Embedded within WeCom
Product overview · 2026-10
Now: Everyone uses their own AI tools
PART 1Why do we need an enterprise-grade AI Harness?From super individual to super organization · Definition · Eight conditions
Why do we need an enterprise-grade AI Harness?

AI has already created super individuals, yet organizations have not grown stronger as a result

Individual capability grows stronger, but AI capabilities remain confined to personal devices: lost upon personnel turnover, unshareable across teams, invisible to managers.

Phase OneAI is a ToolOccasional Q&A, drafting content. Value is fragmented and non-replicable.
Phase TwoSuper AI individualEmployees use desktop Agents and personal assistants to automate entire work segments—efficiency soars, but capabilities remain locked within individuals.
Phase ThreeSuper AI organizationPersonal AI capabilities generated through creation become organizational assets, allocated by role, collaborated via rules, and evaluated by outcomes.
Progressing from phase two to phase three, stuck at five critical breakpoints
01

Assets follow personnel

Useful prompts, skills, and knowledge are stored in individual accounts—lost upon employee departure.

02

Capabilities cannot be reused

Ten people writing prompts for the same task individually; AI capabilities developed by one department cannot be used by others.

03

AI does not collaborate with other AI agents

Each AI serves only one person’s single conversation—no division of labor, handover, or acceptance, so it cannot enter business workflows

04

Data and access permissions out of control

Sensitive data being uploaded into various AI tools without authorization or audit trail—no visibility on who accessed what or how it was used

05

Unable to calculate ROI accurately.

Can only see billing and agent count; cannot see where AI generated results across roles or tasks

PART 1AI Organization Implementation: Redesign how the company operates using AIFrom 1 to 0 · Frontline circle × Backend Foundation · Unified scheduling of people and AI
AI Organization Implementation · Core Principles

From 1 to 0: Not just adding some AI, but rewriting the organization using today’s AI capabilities

“Assuming organizational structure and processes don’t exist, how would a company be designed today using current AI?”

Let AI handle what it can do; let humans focus on judgment, client interaction, and accountability

HAROHumans and AI treated as the same type of resource for unified management
CircleTarget, person, AI, budget handed over to the circle owner
Three rounds of registrationAll employees re-onboard according to new standards
AI InternRe-price 'Experience'
HARO · Unified Scheduling of Humans and AI
Today: Pyramid organization, each using their own AI
Core Diagram · Four-Step Transformation

Traditional functions do not disappear—they become foundational, continuously supporting front-line teams.

Functional Department StaffCircle MembersBP dispatched by the base.AI Employee / Agent

Remain in the foundation layer

    Enter the circle

      Hand over to AI employee

        Current view: Finance → Finance Center (click any module in the diagram above to switch)

        New organization · Four new roles

        Enable AI as a schedulable resource: HARO centralizes 'people + AI'

        CEO / Top Executive: Personally leads the initiative, reporting directly to HARO
        HARO · Human AI Resource OfficeCompilation, capability, cost, performance: humans and AI side by side in one table
        Frontline: several 'circles' = Circle Owner + Member + AI Employee

        In AIPaddle

        Human + AI staffing table
        Member management (WeCom sync) unified with AI Employees and Agents
        AIO Workbench
        Model, permissions, audit trail, and usage are configured centrally in backend
        Tools for Agent Builder
        Generate an Agent and Workflow with one sentence, fixed configuration page maintained
        Product Screenshots
        New unit · Circle

        Replace 'project team' with 'circle': empower those closest to the outcome

        Click to switch comparison

        Where resources come from

        Who makes decisions

        People and AI

        How to evaluate performance?

        Product Screenshots
        Scenario · A day in the life of a circle in WeCom

        No need to learn new systems: just say a sentence in the group chat, and AI Employees and underlying services will activate automatically

        East China Customer Circle (4 humans + 6 AI Employees)Example conversation

        What happened?

        • Assistant automatically routes to Quoting AI Employee
        • Invoke the base finance module 'Gross Profit Forecast' with red-line validation
        • Escalate below-floor cases to Financial BP (L2)
        • Contract initial review conducted through the legal department's contract review Agent
        • Each API call counts toward usage, billed monthly
        Product Screenshots
        New talent · Three-round registration system

        All employees 'stand up again, sit down again'

        First round

        Circle Owner Competition

        First designate team leaders: clearly define who is needed and how AI should be used to deliver results.

        Second round

        Mutual Selection

        People choose circles, circles choose people; headcount planned together with AI Employees

        Third round

        Repositioning

        AI Training Programs, Role Transfers, FDE Talent Pipeline; rules must be disclosed in advance

        Key:Shift hiring criteria from 'seniority and rank' to 'ability to deliver results using AI'.AI Interns can independently build Agents within 3–6 months, paired with senior employees
        New incentive · Measurement

        Bonus pool follows team performance; five metrics determine if actions were correct

        Circle Outcomes

        Revenue, delivery, efficiency

        Circle Bonus Pool

        Finalize based on outcome and distribute down to the community

        Circle Owner Assignment

        Evaluate based on contribution, not job level

        Member + Add-on

        Additional Enhancements for Agent Builder and FDE

        EfficiencyPer capita output
        StructureProportion of Tasks Handled by AI
        SpeedDecision Cycle
        AssetsReusable Agent count
        ParticipationEmployee AI weekly active users
        Product Screenshots
        Platform-driven organizational design implemented within the system.

        Every element in organizational design has a corresponding counterpart in AIPaddle

        Click any row to jump to the corresponding system live view.

        PART 2AIPaddle: Enterprise-grade AI Harness for Organizational ImplementationEntry · Target · Orchestration · Project · Governance · Delivery
        What is Enterprise-level AI Harness?

        Model is the engine; Harness is the full toolkit enabling the engine to serve the organization

        Model is the engine; Harness is the full toolkit enabling the organization to harness it: connect entry points, context, tools, and workflows, apply permissions, approvals, and metrics, and turn every AI capability into enterprise assets

        Entry Point AI follows employees wherever they work: enterprise WeChat groups, workbench, web portal, external systems. Orchestration Personal Assistant · One-sentence generation and invocation Workflow · AI Employee · AI Team · PMO Agent Reusable AI Assets Nine standard AI entities: creatable, editable, publishable, authorizable, reusable, and depreciable Context Knowledge Base · Project Workspace (WeCom Drive) · Organization & Roles Tool: MCP · Business System API · WeCom Interface Governance Define work assignments by position, permissions by role, publish approval, security check, tiered autonomy, full audit trail Measurement. Usage and billing · Work records · Unified KPIs for humans and AI Execution Multi-model access · Multi-tenant isolation · SaaS / Standalone deployment
        Innovation flywheel: how individual-created AI capabilities become organizational assets
        Each full cycle adds another layer to the organization's AI assets
        1Personal CreationEmployees can create Agents, Workflows, and Knowledge Bases with just one sentence in conversation.
        2Capture as objectStandard object stored with source, version, and security check
        3Approve and ReleasePublish to department or all staff; takes effect after admin approval
        4Organization-wide ReuseColleagues directly invoke via WeCom, or assign as AI Employees to roles
        5Measurement feedback.Usage and acceptance pass rate feed back into next-round optimization
        Judgment Standards

        A qualified enterprise-grade AI Harness must meet eight criteria

        The following eight items serve both as selection checklist and as clues for introducing AIPaddle’s modules in the second half of this article: each module corresponds to one or two of these items.

        PART 2Market ComparisonCan Tencent WorkBuddy meet requirements · Cost
        Market Comparison · Tencent WorkBuddy

        WorkBuddy is a powerful individual-level super tool; organizational-level Harness is something else entirely

        WorkBuddy empowers individuals; AIPaddle strengthens organizations. Below is a summary based on publicly available information (as of October 2026).

        VersionPriceQuota & Details
        Personal · Standard Edition99 yuan/month2,000 base + 2,000 bonus points per month (effective from July 1, 2026; previously priced at ¥58)
        Personal · Advanced EditionRMB 199 per monthHigher points quota
        Personal · Premium Edition999 yuan/monthHighest personal quota
        SaaS Enterprise EditionRMB 198 per person per monthRMB 2,376 per person per year; 2,000 points per person per month, shared across team; minimum purchase: 1 seat (effective May 15, 2026; previously RMB 78)
        Dedicated Cloud Enterprise Edition¥316 per person per monthRMB 3,792 per person per year; dedicated VPC; minimum purchase: 100 seats (previously RMB 158)
        Private DeploymentPre-sales consultation.Third-party materials billed as project-based annual fees, ranging from hundreds of thousands to millions of RMB
        Credit Add-on¥100–¥9,2002,000–200,000 points, valid for 6 months; point consumption varies by model, token count, and task complexity.

        Annual cost estimation.

        Estimated annual cost.¥237,600Subscription fees paid annually per seat; additional purchase required when credits are exhausted
        Cross-reference eight conditions
        WorkBuddy excels atEmpower everyone to be stronger: handle local files directly on desktop, produce high-quality office deliverables like reports, PPTs, spreadsheets—ready-to-use out of the box with rich ecosystem and model resources
        Organization-level requirements still neededCapabilities are accumulated into organizational assets that can be published and authorized; humans and AI collaborate by role and are embedded into project workflows; executed and evaluated according to enterprise policies; data and models fully owned by the enterprise
        RecommendationThe two approaches are not conflicting: individuals can continue using their preferred AI tools; organizational AI assets, workflows, permissions, and metrics are managed within the company’s own Harness.

        Sources: Tencent Cloud 'WorkBuddy Enterprise Billing Overview', Tencent Cloud Developer Community 'WorkBuddy Enterprise Limited-Time Offer', Cailian News / East Money 2026-04-28 billing adjustment report, WeiYi Network 2026-06-17 personal plan price adjustment report, Real Intelligence 'What’s the Difference Between WorkBuddy Personal and Enterprise Plans?', Eigent 'WorkBuddy AI Review (2026)'. Pricing subject to real-time quotes on Tencent Cloud official website.

        PART 3Key Modules and Business Scenarios of AIPaddleEach module corresponds to a Harness condition
        AIPaddle · Enterprise AI Harness

        AI Harness embedded in WeCom

        WeCom serves as the entry point, WeCom Drive stores documents and records, while AIPaddle handles the production, accumulation, collaboration, and governance of AI capabilities.

        Capability Mind Map (click any branch to jump to the corresponding section)
        AIPaddleEnterprise-level AI Harness Entry Point WeCom Group Chat @PMO Dashboard · Mobile App Personal Assistant Copilot Page Help Nine types of AI entities Agent · AI Employee Workflow · Scheduled Job Skill · Knowledge Base MCP · API Plugin Dynamic Orchestration Generate an Agent with one sentence AI Orchestration of Workflows Fixed configuration page maintenance Project management PMO Plan · Gate Risk · Issue Log Daily report · Weekly report · KPI Rules and automation Governance Access Control Approval Center AI Operation Logs Usage · Monthly Billing Delivery & Integration SaaS · Independent Deployment ERP / SAP Integration Enterprise-owned models Project Workspace in WeCom Drive
        Product Screenshots
        System Architecture · Business Perspective

        From a single sentence in a group chat to a line in WeCom Drive

        Five-layer architecture. Top layer: WeCom used daily by everyone; middle layer: AI execution layer; bottom layer: project spaces and governance. Each layer interacts only with adjacent layers.

        Entry Point WeCom Project Group@PMO · Task Assignment · Reporting · Approval WeCom Workbench (mobile)Dialogue · View · Start/Stop Personal Assistant (Web Version)Create AI · Voice · Drag Files with one sentence Project page / management backendTeam · Template · Permissions · Approval AI Execution PMO AgentInspection · Follow-up · Escalation · Issue analysis · Meeting coordinationAcceptance · Approval Routing · Daily/Weekly Reports · KPI Personal AssistantUnderstand → Build → Schedule → AssignName invocation · Auto-assignment · Generate report AI Employee 🤖Attached to position · Receive task · Deliver outputHuman responsible person verification capabilities Nine types of AI entities Application: Agent · Workflow · AI Employee · AI Team; Foundation: Knowledge Base · Skill · MCP · API · Extensions; + Scheduled Job · WeCom Push Node. Project Workspace Project Workspace in WeCom Drive11 standard directories · Originals & deliverablesAutomatically extract file content into AI-readable format Online spreadsheet = Original sourceExecution plan · Risk Ledger · Issue LedgerChanges made by people in the sheet are bidirectionally synchronized every 5 minutes Knowledge BaseArchives · Team · System Automatically IngestedAnswer with proper sourcing Governance Access & Security Assign tasks by position, define permissions by role, department data access, approval center, tiered autonomy (direct action if reversible, approval required otherwise), multi-tenant isolation, full audit trail
        AIPaddle Personal Assistant · Create any AI object via chatKey

        Say something human-like; have nine AI agents built as needed

        State your requirement; assistant first drafts plan, confirms with user before creation; created objects can be edited again and published to specific individuals or departments

        Business personnel say:

        Assistant's creation plan (by sequence) ①→⑩）Already Live

          ① UnderstandIntent is categorized into only three types: chat, create, create and schedule; when information is insufficient, the system asks one question at a time and provides clickable options.
          ② Create pair.Build in dependency order: create referenced items first; check before creating—reuse existing ones, automatically request permissions if lacking access
          ③ Scheduled«Every day at 8:00» parsed and echoed back by code, displayed on «Scheduled Job» page marked as «AI Auto-created»
          ④ DelegationEvery step is visible; reasons are recorded for reuse, skipping, or degradation; the entire batch can be undone with one click
          Full lifecycle of object
          Objects created by assistant follow same lifecycle and enter same editing page as manually created objects
          01Create dialogueGenerate a creation plan with one sentence—confirmed and then deployed in sequence
          02Security MonitoringEach save triggers synchronized scanning: hardcoded key interception, injection protection auto-repair
          03Trial ConfirmationFirst run a dry test—preview only, no sending. Only enable scheduled posting after confirming 'Yes, meets expectations'.
          04Private access availableCreated by the individual immediately available; others cannot see it
          05Secondary editingEdit prompt, switch knowledge base, adjust workflow in corresponding editing page; alternatively, say 'Wrong, need to change' during conversation.
          06Application PublishedSelect enterprise public spaces / designated individuals / designated departments, submit for approval
          07Approval AuthorizationOrganization administrator reviews first, then may forward to department AIBP for final approval; approval becomes effective immediately for the authorized person.
          08Deactivate / WithdrawPublished objects must be offline before deletion; higher-level dependencies will be flagged accordingly.

          Where is secondary editing located?

          Already Live

          Agent → Agent Management & Orchestration Page; AI Employee → AI Employee Workbench; Workflow → Canvas (with version history); Knowledge Base → Knowledge Base Details; Skill → My Skills; MCP → MCP Edit Page; Extensions → Extension Settings Panel. Direct conversation debugging is available on the right side of the edit page.

          Who should this be published to?

          Already Live

          Three publishing targets: enterprise-wide public space (all staff), authorized to designated individuals, authorized to designated departments (including sub-departments). All require approval. Authorized users see and use resources in Resource Marketplace; use «Type#Name」invocation

          Who approves

          Already Live

          Two-stage review process: First, organization administrators conduct initial review; if unclear, escalate to AIBP for final decision by the applicant’s department’s AIBP. The Approval Center filters by pending / approved / rejected, listing 'this time authorized to' and 'currently effective'.

          Usage Example
          Product Screenshots
          In-depth introduction to the nine types of AI entitiesKey

          Four types of application objects perform tasks, five types of foundational objects provide support

          All AI capabilities are limited to nine standard object types: four used directly, five referenced or responsible for integration; reference relationships are enforced by the system, traceable and governable.

          Business Organization Layer AI TeamMember = AI Employee / Agent, orchestrated through workflows AI EmployeeHigher-level Agent with defined responsibilities AI Execution Layer WorkflowMulti-step orchestration · Push to WeCom node AgentModel + instruction + resource, minimal execution unit Capabilities and knowledge layer Knowledge BaseRetrieve corpus SkillBusiness Method / SOP MCPStandard Tool Services APIExternal system interface Integration Layer Extensions (inbound-only)WeCom Robot · External System API · Webhook → Call published Agent / AI Employee / Team Enforced by code-based prohibition relationships Standard Agent cannot be assigned as another Agent AI Employees cannot be assigned to other AI Employees Teams cannot be nested within other teams Skill cannot rely on Workflow Workflows cannot be registered as tools Multi-Agent collaboration must go through workflows Permissions can only be narrowed down layer by layer, never expanded. Dependent objects cannot be permanently deleted—they can only be taken offline
          Product Screenshots
          Dynamic Agent · Dynamic WorkflowExclusive

          Speak one sentence to generate an Agent and workflow, then maintain it in fixed configuration page

          First state requirements; AI generates configuration and fills it into fixed configuration page. Users make minor adjustments and test conversations on the same page, then publish through approval process.

          1

          State Requirement

          Describe what to do using natural language

          2

          Generate Configuration Patch

          Structured JSON, field-by-field verifiable

          3

          Real-time entry into configuration page

          Left-side fixed fields updated item by item, highlighted

          4

          Manual fine-tuning · debugging

          Edit on the same page—right side instantly shows live chat preview

          5

          Draft → Approval → Publication

          AI can only draft; it cannot publish content on its own

          PromptYou are the Lease Ledger Reconciliation Assistant, responsible for comparing lease ledgers against on-site reconciliation forms…
          Variableproject_code · check_date
          Brain SettingsFunction Calling (Tool Invocation)
          Knowledge BaseProject library - Beijing Tongzhou Dachuan Project · Asset Management Methodology Library
          Tool (Skill)WeCom Drive read/write · Asset Management Calculator
          Recommendation-related questionsReview current lease ledger and list discrepancy accounts...
          Generate Main Controller (Natural Language)
          Create a lease ledger reconciliation assistant: read lease ledger and on-site verification forms from Project Workspace, compare unit by unit, output discrepancy list, and register findings in Risk & Issue Log.
          Configured: System prompt written; added two variables: project number and verification date; linked project knowledge base; connected WeCom Drive read/write and calculation tools. Each generation explicitly notes changes made and cost incurred.
          Sample animation: fields and workflow match those tested on 2026-10-10; only resources authorized to current user will be matched.

          Dynamic Agent

          • Agent Management → 'AI Help Me Build': Generate draft with one sentence
          • Right-side 'Generate Master': describe need → auto-generate or fill left-side config and match authorized resources
          • All prompts, variables, models, knowledge bases, Skills, and plugins are placed in fixed fields

          Dynamic Workflow

          • Workflow canvas → "AI Orchestration": add a node or modify branches with one sentence
          • Canvas updates in real time; version can only be saved after passing automatic health check
          • Can be invoked by scheduled jobs, WeCom bots, and external APIs

          Four lines of defense

          • Structural validation: patch with invalid format rejected outright
          • Permission boundary: only matches authorized resources
          • State Machine: AI only drafts, cannot publish
          • Audit Log: Every modification is traceable and reversible
          Already LiveAI Assistant, main control creation, and AI orchestration are now live in the Pinqi production environment
          Product Screenshots
          WeCom API Management · ExtensionKey

          One page fully configures all WeCom integrations

          One page configured with full WeCom integration: capability overview clear at a glance, credentials encrypted by enterprise

          Contact list synchronizationContact list sync Secret (read-only)

          Organization structure and members automatically sync at 8:00 AM and 8:00 PM daily, or can be manually synced immediately; onboarding/offboarding takes effect automatically.

          Message push · Passwordless loginSelf-built application AgentId + Secret

          All workflow nodes "pushed to WeCom", PMO proactive push, and dashboard seamless login use the same credentials

          Group robot enters stationSmart Robot BotID + Secret

          Persistent connection mode, no need for domain registration; bind the desired Agent / AI Employee / AI Team in the 'Extensions' section.

          Extension: The sole entry point for external systems calling AI

          WeCom Bot

          Already Live

          In group @Robot → routed to extended bound entity → streamed back into group chat. Registered robots can be reused directly from dropdown (one robot serves only one extension at a time).

          External system API

          Already Live

          Open conversation interfaces to official website, mini-program, and business systems: API Key stored only as hash, plaintext displayed once only, revocable anytime; rate limiting by both Key and IP dimensions, source domain whitelist enforced.

          Webhook trigger

          Already Live

          External system pushes a message, asynchronously triggers Agent or workflow; supports IP whitelist, rate limiting, 64KB limit, results recorded in trigger history

          External calls still subject to data isolation constraints; rejected upon deactivation; individual call tracking maintained

          Usage Example
          Product Screenshots
          WeCom + WeCom Drive · Project WorkspaceKey

          Throw a document into the group, and AI answers immediately

          Submit files to platform: automatic categorization, conversion to online documents, extraction of AI-readable text; relevant materials to be added to Knowledge Base

          Inspect WeCom Drive every 5 minutes; files manually dragged into WeCom Drive are also detected
          1 · Entry

          Upload in group or on page

          Leasing Lead sends meeting minutes to the platform or directly drags them into the project workspace.

          20260925_Leasing Meeting Minutes.docx
          2 · Determine Direction

          Automatically determine location placement

          File documents according to the rules: meeting minutes go to '04-Meetings & Decisions/Special Meeting Minutes'; uncertain cases go to '99-Pending Classification' for confirmation.

          04-Meetings & Decision Making/Special Meeting Minutes/
          3 · Digitization

          Convert to online document

          Original documents moved to 'Processed Originals'; body content converted into collaborative, editable WeCom online documents that allow AI to write back.

          Online document · Original version archived
          4 · Refinement

          Convert into AI-readable text

          AI always reads the latest file version; archival, team, and policy documents are simultaneously ingested into the Knowledge Base.

          .md Extraction · Effective within 5 minutes
          5 · Put It Into Use

          Q&A, task assignment, archiving

          When a member asks "Where are we with the anchor tenant terms?", PMO responds based on meeting minutes and cites sources; action items from minutes are added to the task ledger.

          Source: 04-Meetings & Decisions/…

          Organization Sync · Passwordless Login

          Already Live

          WeCom contact list = account; onboarding/offboarding auto-syncs; ready to use immediately

          Creating a project automatically establishes a group and workspace

          Already Live

          After selecting template, automatically create WeCom Drive space, 11 standard directories, 12 essential documents, and register tables, and bind to project group

          Task assignment in group, real @ Assign to person

          Already Live

          «Assign to Leasing Manager: ...» written into execution plan as a single sentence and @ Assign to person; if no deadline set, follow-up will be triggered

          Online spreadsheet bidirectional sync

          Already Live

          The execution plan table is the single source of truth for task status. When people make changes in the table, the project page and PMO read back the data every 5 minutes.

          Direct management of WeCom Drive within the platform.

          Already Live

          Upload, create, rename, move, delete files directly within project page—no need to switch back to WeCom to locate files

          Document inbox · AI sorting

          Already Live

          Disorganized documents sent by Party A are automatically split and added to corresponding files and sections by AI; revisions and traceability are supported

          Personnel access and permissions follow people

          Already Live

          Removing project automatically revokes WeCom Drive space permissions and performs read-back verification to ensure removal is complete.

          WeCom Workbench mobile version

          Under construction

          Interact with, view AI objects, and start/stop scheduled jobs via mobile; mobile use only, no configuration required

          Conversation Archive Integration

          Under construction

          Group chat history stored for use in quoting and meeting minutes; enterprise must enable WeCom conversation archiving

          WeCom capability boundaries – we clarify these upfront

          • AI can only read and write documents and spreadsheets created within the platform; project ledgers must be uniformly established through the platform to avoid this issue.
          • WeCom Drive space created by the platform; members can only view and download files in WeCom; uploads must be completed via project page or assistant.
          • WeCom does not return the editor of online documents; Smart Sheets can identify changes per row, making it the authoritative source for team member work records.
          • Conversation archive requires separate purchase by enterprise and retains data for only 5 days; historical data prior to activation cannot be retrieved.
          Usage Example
          Product Screenshots
          Enterprise-level + Project-level WeCom Drive Space · DRIVE-2New

          Each enterprise has one shared workspace, and each project has its own project space, with outputs automatically filed in the correct location.

          Project files go into the project workspace; AI object outputs go into the enterprise-wide public space—anyone who can use the object can view its output

          In planningRequirement approved (Oct 8, 2026), development pending. Three demo options below; permissions derived from real user tests on test WeCom enterprise
          Where does a file land? View decision process by selecting scenario
          Evaluation order: top-down, stop at first match
          1. Location specified during invocationSpecified space/folder or assistant workspace already bound to folder → Use specified
          2. Belongs to a certain projectEnter the project’s own project workspace (status unchanged)
          3. Other outputsEnter the object’s folder inside the enterprise-level workspace
          4. Unable to determine who initiated the requestEnter your own "04-Personal Assistant / Name"
          No longer revert to the most recently created project space.
          AIPaddle Public Space (automatically created by platform, zero configuration)
          AIPaddle Public Space/
          01-Workflow/
          Daily Real Estate News Digest
          Contract Review /2026-10/
          02-Agent/
          Investment Assessment Assistant/
          03-Skill/
          Meeting Minutes Compilation/
          04-Personal Assistant/
          Perry Xu/
          Zhang San/
          Project-level workspace (one per project, current status unchanged)
          Beijing Tongzhou Big Ship Project/04-Meetings & Decisions/

          Folder created only upon first output of object; folder name updates with object rename; folder retained after object deletion, visible only to administrators.

          Who can view: tested through three rounds of real-user testing on WeCom
          Test team members' WeCom · WeCom Drive

          Conclusion: In WeCom, adding someone as a space member grants them visibility into the entire space, and there is no interface to disable inheritance. Therefore, the solution is set asDo not add members—only assign folder-level permissions, access directly from platformVisible that human access rights follow object permissions—visible only to those it's published to, and revoked along with the folder once publication is withdrawn. Due to WeCom restrictions, folder authorization only supports 'download-only'.

          Add new 'Public Files' tab (example): switch identities to view visibility scope
          Who is viewing from what perspective?
          No permission; please contact the creator for sharing

          Automatically create spaces and folders

          In planning

          Upon connecting WeCom Drive, the "AIPaddle Public Space" and four primary folders are automatically created;⑦ Remove spaceid and 'Default Directory' input fields; retain only 'Rename'. Existing enterprises continue using original spaces; legacy files remain unchanged.

          «Public Files» page

          In planning

          New sidebar entry visible to all members. Authorized users can expand to view files and jump to open them in WeCom; unauthorized users see grayed-out locked icons without file counts. The Personal Assistant folder only displays files owned by the user.

          Include links in the notice

          In planning

          After output completion, file links are directly attached in WeCom notifications—click to view; quick access to "output files" added to workflows, agents, and skills pages

          System acceptance test cases (15 items, tested individually with screenshots after go-live)
          #ScenarioShould be visible.Status
          Product Screenshots
          Project management · Full lifecycle of PMO AgentMost critical

          Rules are embedded in policies; execution is delegated to PMO Agent

          One PMO Agent manages one project group: each step recorded in the ledger, every notification includes regulatory basis. Select a stage to see what happens in the group.

          Project group
          Usage Example
          Project detail page: five tabs per project (Pinqi production environment launched October 10, 2026)
          01

          Project archive

          • Basic records (based on 'Project Initiation Document')
          • WeCom Group Management · Chat history archived
          02

          Project plan

          • A1–A6 Phase · Execution Plan
          • Deliverables and gate review for this phase
          03

          Team Management

          • Position · Line of Business · Responsibilities · Team Members
          • Assign AI Employees to roles
          04

          Document management

          • Project workspace (WeCom Drive 00–09)
          • Project knowledge base
          05

          Collaborative Management

          • AI Collaboration · Risks & Issues · Meetings & Minutes
          • Reporting and performance evaluation (daily / weekly / KPI) · Rules and automation
          Product Screenshots
          Project team = Human employees + AI employees

          AI Employees and humans work the same role with identical responsibilities and performance evaluations

          Assign AI Employees to roles with a designated human responsible party; AI is prioritized for task assignment, but delivery is validated by the human

          Under construction First batch under development

          Legal Contract Lead

          Contract review, clause risk, contract ledger
          🤖Contract reviewerAI · Responsible Party: Legal & Contracting Lead
          HumanLegal Contract LeadIn Progress 3

          Leasing Lead

          Business Format Planning, Key Tenant Negotiations, Leasing Progress
          🤖Leasing Weekly Report AssistantAI · Responsible Party: Leasing Lead
          HumanLeasing LeadIn Progress 5

          Renovation Project Lead

          Renovation design, construction permit, fire safety, and final inspection
          HumanRenovation Project LeadIn Progress 4

          Finance / Funding Lead

          Annual budget, NOI calculation, rent collection.
          🤖Rent collection verification officerAI · Responsible Party: Finance Lead
          HumanFinance LeadIn Progress 2
          On-time completion rateHuman/AI equivalent metrics
          One-time acceptance rateAI Must Be Verified
          Re-work countCumulative return
          Average delivery duration.Issue → Complete
          1. 01Project manager in group: "Assign to Legal & Contracts role: review the new lease contract in 05-Contracts & Business"
          2. 02PMO Matches Roles to People: Legal & Contract Role Has Registered AI Employees → Assigns to Them 🤖Contract reviewer，@ Its responsible party
          3. 03AI Backend Task: Review contract files flagged in read-only tasks, generate review comments, and write them back into the project workspace
          4. 04Task enters 'Pending Acceptance': '🤖Contract reviewer has delivered T-021,@Legal Contract Lead: Please verify
          5. 05Responsible person replies to "T-021 Rejected: Supplement analysis of rent-free period terms" → AI revises with feedback (up to 2 rounds)
          6. 06Responsible person replies to "T-021 Approved" → Completion achieved; on-time completion rate, first-pass success rate, and rework count are included in the AI Employee's KPI
          Usage Example
          Product Screenshots
          Copilot · Self-IterativeNew

          Every page has Copilot; every user issue feeds into the product’s next improvement cycle

          Copilot is available on the right side of every page; unanswerable queries can be instantly converted into issues, entering a self-iterative closed loop: 'AI repair, human verification'

          COPILOT-1

          Page helpUnder construction

          Recommend questions by page; include source in answers; if not found in manual, state directly; display beside title ⓘ Explain data definition;⌘J Trigger; Full-screen edit page automatically collapses. Focuses solely on how to use, does not access business data or perform actions on your behalf.

          COPILOT-2

          Ask questionsUnder construction

          Description + up to 3 screenshots; click "Not Resolved" to instantly convert into a ticket; enter risk log with AP-xxxx ID; max 10 per person per day.

          COPILOT-3

          My previous inquiriesIn planning

          Pin three entry points (can't operate / found bug / new requirement); view processing status and replies; status changes push notifications to WeCom

          Self-iterating closed loop

          1Three entry points

          WeCom Online Spreadsheet · Copilot asks questions · Operations Console

          2Issue log

          Unified ID AP-xxxx, deduplicate and categorize

          3AI conducts analysis first

          Identify root cause, provide remediation plan, and assess impact scope

          4Responsible Person Confirmation

          Proceed only after approval; otherwise reply to the questioner

          5Automatically develop

          Independent branch development, self-testing, and PR submission

          6Manual consolidation and release

          Production branch hard gate; notify result to the questioner

          Orange two-step processes must be completed manually: AI does not directly modify live systems

          Interface illustration (clickable; actual screenshots will be added after Pinqi’s official environment sync)

          MonitoringiWhere do the monitoring page data come from?
          Real-time monitoring of enterprise AI usage (data sourced from invocation logs and security audits; cost estimated based on token unit price)

          Today's usage volume10
          Token Consumption0.04M
          Today's estimated cost¥0.04
          Agent published25/26
          Skill published21/22
          Invocation Success Rate99%

          Left panel shows the monitoring page for Pinqi tenants; right-side Copilot answers only 'How to use the platform'—it does not access business data or perform actions on your behalf.

          CopilotPage Help · Currently Viewing: Monitoring
          You might be asking
          Answer only platform usage questions; do not access business data or perform actions on your behalf
          Encounter issues or have suggestions? Submit a question
          Ask questions
          Sent to AIPaddle product team – do not fill in passwords, customer privacy information, etc.
          + Add screenshot (optional, up to 3)
          Under constructionCOPILOT-1/2 Test Environment Live; Pinqi Production Environment Pending Sync; COPILOT-3 Under Planning
          Short video montage · Auto-generation from project materialsNew

          Automatically generate short videos from project materials: review draft first, then finalize; one video daily

          Media uploaded to project WeCom Drive, system automatically analyzes; script reviewed before editing, final cut produced daily and archived to project WeCom Drive (see "Short Video Mashup")

          System performs hourly inspection of media, analyzing only newly added or modified videos
          01

          Specify Media Library

          On the top-right corner of the project, the "Media Library" automatically creates "Original Media" and a peer-level "AI Delivery Media"

          02

          Upload assets

          Upload in 'Materials Management' or transfer to '99-Pending Categorization'; system detects video and auto-transfers it.

          03

          Analyze content materials

          Scene transitions precise to frame level; one frame analyzed per second within each scene, recording visual content, original subtitles, and audio.

          04

          Submit requirements

          In Personal Assistant, clearly state: topic, audience, duration

          05

          Five confirmation cards

          Confirm sequentially: content breakdown → template (default B version) → voice tone (49 options available) → background music → script.

          06

          Output production and rework

          All revisions—editing, changing visuals, adjusting parameters, switching themes—can be discussed within the same conversation. Each revision generates a new version, with previous versions available for review.

          07

          Sample footage approved

          First finished video must be reviewed by project lead, then click "Sample Approved"

          08

          Daily Scheduling

          Script finalized and verbatim draft completed by 18:00; sent to approver; approved video automatically generated at 08:00 the next day.

          09

          Archiving delivery

          Five file types—final video, cover image, subtitles, transcript, source list—are grouped under "AI Deliverables"

          Three rules

          • Review script first, then edit: no voiceover or video production until marked 'approved'
          • Sample footage is only valid after human review; thereafter, daily script reviews only
          • Unconfirmed drafts remain stored indefinitely without loss or replacement

          Final video specifications

          • Portrait orientation, approximately 60 seconds
          • Remove voice modulation, single narrator, 1.2x playback speed
          • Background music placed beneath narration, single-line subtitles synchronized with voiceover
          • Use the same template style for intro and cover design
          Already LiveProject entry, media library, daily short videos, templates, and voice settings have been launched in the Pinqi production environment; music library now hosts 97 commercially usable tracks; once the designated media library is set for the Dachuan project, content can be produced following this workflow
          Product Screenshots
          Full View of Business Scenarios

          From a functional perspective: what does Harness do on each business line?

          Same Harness, different AI assets accumulated across each business line (e.g., commercial real estate and project-based enterprises).

          Product Screenshots
          Knowledge Base · Mobile · Content Production

          Most frequently used daily capabilities

          Original Format Preview and Collaborative Editing of Office Files

          Under construction

          Word/Excel/PPT files in knowledge base open in original format, support multi-user real-time editing; after saving, automatically re-uploaded; shared knowledge bases automatically revoked and re-approved

          WeCom Workbench mobile version

          Under construction

          Mobile-only mode: interact with assistant (voice, photo, QR code), view objects, start/stop scheduled jobs, view to-dos; objects fully consistent with desktop version.

          Name invocation · Word report

          Already Live

          Write "agent" in WeCom#Name」can be invoked directly; when reporting or answering requires lengthy content, send text first, then full .docx file and a 24-hour download link

          Image-to-Text Generation

          Already Live

          Daily, retrieve images from WeCom Drive photo library by category, use visual model to analyze, generate Word document with 【Figure N】insertions, and push

          Short videos produced by theme

          Already Live

          Media automatically identifies shots, visuals, and voice; generates transcript by theme, reviewed before final edit, one video produced daily and archived to project WeCom Drive (see "Short Video Mashup")

          Voice · Drag-and-Drop · Presentation Theme

          Already Live

          Convert spoken words into text input; drag files as attachments, drag folders as reference materials; upload Word file and select 13 presentation themes for demo

          Usage Example
          Product Screenshots
          PART 4Technology & DeliveryTechnical development · Usage and billing · ERP integration · SaaS architecture · Standalone deployment · Security · Roadmap
          Detailed technical development

          Full-stack TypeScript monolith, Postgres as foundation, database-level isolation as fallback

          Full-stack TypeScript monolith, browser does not connect directly to database, row-level security for each business table as fallback

          ≈RMB 208,000Execute TypeScript code
          148Database Migration (with rollback scripts provided)
          81Enabled row-level security on Table A with 132 policies
          764Automated Test File
          ≈80Permission Action, default denied
          ≈40copies of architectural decision records
          Expand: Technology selection and layering
          FloorSelection
          LanguageTypeScript 5.7 (strict mode), unified frontend and backend
          FrameworkNext.js 16 (App Router) + React 19; Server-side Components Direct Data Output
          InterfaceTailwind CSS 4, Radix / shadcn components, Recharts dashboard, ReactFlow workflow canvas
          BackendNext.js API Routes / Server Actions, Monolithic Architecture, No Independent Backend Service Built
          DatabaseSupabase: PostgreSQL 17 + pgvector vector search; Auth authentication; private object storage
          Large Language ModelAlibaba Cloud BaiLian Qwen (OpenAI-compatible protocol), enterprises can configure other compatible vendors
          File processingOfficeParser / UnPDF for reading; docx / exceljs / pdf-lib for generating; OnlyOffice for native-format collaborative editing; FFmpeg for video processing
          IntegrationWeCom long connection robot (WebSocket), WeCom API, MCP client, SMTP
          DispatchCroner parses cron expressions in Beijing Time; server triggers scheduling entry every minute
          Validation and TestingZod for unified input validation; Vitest for unit/integration testing, Playwright for end-to-end testing
          ExecutionNode.js 22，pnpm；Ubuntu + pm2 + nginx（TLS）
          Browser / WeComClient-side
          Only call internal APIsNever connect directly to the database
          Edge middlewareEvery request must go through
          Login guardian12-hour login-free session / 24-hour usage capAudit TrailMobile / Desktop split
          API RoutingEntry Authentication
          Check permission matrix by actionDeny by DefaultAutomated testing scans for missing authorization
          Data LayerThe sole layer accessing the database
          First participant fixed as request contextRuns only on the service side
          PostgreSQLEnterprise-level security fallback
          Each row includes an enterprise IDCross-enterprise query rejected by database
          How one request flows: within the group @Robot query
          1 IntegratedLong connection received in WeCom @ Find the extended object bound to the messageInstrumentation maintains persistent connection
          2 IdentityInitiated by trusted session: "Who, which company, what role"Do not trust frontend parameter transmission
          3 AuthenticationCheck permission matrix by action; default to denyrequirePermission
          4 · Data RetrievalData layer includes user identity queries, row-level security enforced simultaneouslylib/data
          5 · SearchIssues vectorized, similarity search via pgvector, integrated into prompt1536-dimensional
          6 Fallback OptionsWhen retrieval is empty, enforce 「Prohibited from Fabrication」 constraint via codeNot reliant on prompts
          7 GeneratedModel outputs streamingly, refreshing group messages in real-time during generationQwen
          8 Posting to LedgerInvocation logs, token usage, audit trail written to databaseFailure does not block response
          How to integrate with the large model
          AI InvocationDialogue · Embedding · Visual · Voice Model parsingObject configuration → enterprise default → platform default Enterprise-assigned vendorsOpenAI-Compatible: Qwen / DeepSeek / Zhipu / Private vLLM Platform key (SaaS).Bailian Qwen, pay-per-use billing AIPaddle Model Gateway (Independently Deployed)Client machine holds revocable token only Alibaba Cloud BaiLian · Qwenqwen-plus / max / vl-max · asr / tts

          Unified Exit

          All AI calls routed through a single module: conversation, streaming, tool invocation loop, vector embedding; default model is qwen-plus; Agents may independently select max/turbo/vl-max models

          Enterprise-configured models

          Configure OpenAI-compatible endpoint and API key (encrypted storage) under 'Model Provider'; set default slots for conversation, embedding, re-ranking, speech recognition, and synthesis; configuration errors automatically revert to platform defaults.

          Usage & Quotas

          Each API call logs input/output tokens, model used, and source; estimated cost calculated by platform unit price; if enterprise usage exceeds quota in the past 30 days, system triggers circuit breaker

          Audio & Visual

          Speech recognition: qwen3-asr; speech synthesis: qwen3-tts (30 voice styles); image understanding: qwen-vl-max; throttling handled via queuing and retry with backoff

          Retrieve online information

          Invoke Qwen-native online search to answer with source links; if retrieval fails, downgrade to standard conversation and clearly state the limitation

          Vector embedding

          text-embedding-v4, 1536-dimensional, cosine distance; chunk size capped at 512 tokens to ensure retrieval precision

          Persistent Services and Background Tasks

          • WeCom long connection: established with service process startup, connects automatically based on published extensions and reconnects if disconnected
          • Scheduled scheduling: scheduled jobs, PMO inspections, organization sync, document polling, and session archiving pull are triggered every minute by server.
          • Personal Assistant creates tasks: backend orchestration execution with state persisted in database; service restarts are resumed via heartbeat mechanism, ensuring no task loss.
          • Video processing: ffmpeg runs in a restricted resource group (CPU and memory capped), ensuring no impact on online services.
          • Office Collaboration: Standalone OnlyOffice Document Service, Signature-based Access, Auto-reimport after Save
          • Conversation Archive Sidecar: deployed on enterprise-owned servers, decryption private key remains exclusively on enterprise machines.
          Product Screenshots
          Usage and billing · BILL-1New

          Each AI operation's cost is visible immediately, traceable afterward, and billed monthly

          Immediate visibility of token cost for each AI operation; individual access, organizational oversight, monthly billing

          Knowledge Base · Beijing Tongzhou Dachuan Project (Demo)
          Click an action, wait ~2 seconds, check bottom-right corner.
          01-Team & Responsibilities__Job Description.md
          Real estate news summary (workflow)
          Personal Assistant

          Demo: token and amount for three operations are sourced from test environment data recorded on October 7, 2026.

          Visible immediately

          Small text at bottom-right corner, non-intrusive, does not pop up or auto-disappear; next operation will overwrite it

          Two-tiered Query

          «My Usage» shows only personal usage; «Organization Usage» displays every round for each member, supports filtering and export

          Clearly state what was done

          e.g., "AI-generated Agent · Name"; failed iterations marked as "Failed"; scheduled jobs display creator

          Auto-generate monthly bills

          Settle billing monthly based on Beijing Time, with statements issued at the beginning of each month; three tables (by member, model, scenario) are exportable

          Billing Methodology

          • Amount = Model cost + platform service fee
          • Model cost based on official list price, updated daily
          • If price cannot be found, mark model as 'Not Priced'—do not estimate.
          • Billing continues even if invocation fails; price changes only affect future invocations
          • Enterprise-provided model key: no model fees charged, only service fees apply.
          Already LiveMy usage, organizational usage, and monthly billing have been live in the Pinqi production environment; the small text showing consumption in the bottom-right corner and mobile app usage are based on test environment measurements
          Product Screenshots
          ERP Integration · SAP and OthersKey

          Integration with SAP and other ERP systems: how data enters and exits, and how identity is synchronized

          Five universal integration channels to connect with ERP systems like SAP, four of which are already usable; click any channel to view documentation.

          Principles　Use API only, no direct database connection · Read-only preference · Write-back requires approval · Credential encryption · Full audit trail throughout calls
          Authentication: One enterprise identity effective across three systems
          Enterprise Identity Provider (IdP) SAP IAS / IPS · Entra ID · WeCom Address Book AIPaddleUser · Department · Role WeComuserid · address book SAP S/4HANAPFCG Role · Organizational Level AIPaddle: Which identity is used to invoke SAP? Technical User / User Identity Propagation Employee ID / Email ↔ WeCom User ID ↔ SAP User
          1. ① Single Sign-OnWeCom password-free login is live; SAML/OIDC integration with enterprise IdP is under planning.
            In planning
          2. ② User and Organizational SupplyWeCom contact sync now live; IPS/SCIM push integration planned
            In planning
          3. ③ Technical user accessing SAPCommunications user calls OData; data scope narrowed by AIPaddle per department; full audit trail maintained
            Already Live
          4. ④ User Identity PropagationOAuth2 SAML Bearer; SAP Authorization Based on User Role, Also Required for Write-Back
            In planning
          5. ⑤ Permission alignment and auditSAP organizational level → department data scope; audit logs retained on both sides.
            In planning
          Product Screenshots
          SaaS Multi-tenant Architecture

          One platform serves multiple enterprises, data isolated at row level

          One enterprise per tenant: data isolated at row level, credentials encrypted by enterprise

          Platform super admin (independent whitelist, unrelated to internal enterprise roles).Tenant enable/disable · Operation dashboard · AI usage · Platform-level MCP registration · Model price maintenance Tenant AOrganization & Department (WeCom synchronized)Role: Administrator / AIBP / UserWeCom Configuration · Model Provider · MCP Credentials (encrypted)Nine types of entities · Projects · Knowledge Base · Quotas Tenant BSame structure, fully isolated dataCannot see any objects or credentials of A One account can belong to multiple enterprisesOwnership relationship determines 'who can enter which venue'Current enterprise decision on 'identity used for operation'Switching controlled via interface; cannot be self-assigned PostgreSQL row-level securityEach row includes an enterprise ID · Strategy reads 'current enterprise' · High-privilege keys appear in only one file and must explicitly include enterprise filtering

          Tenant lifecycle

          Already Live

          Platform super admin activates and deactivates tenants; deactivation takes effect immediately (members lose access instantly), and reactivation is possible; each tenant has assigned model and storage quotas.

          Permission Configuration Panel

          Under construction

          Role×Page (invisible / read-only / full), role×Data Scope (Entire Organization / This Department and Subordinates / This Department / Myself), Department×Unified configuration for the three tables of business data.

          Estimate invoice based on volume

          Already Live

          Estimate costs based on unit price at time of invocation; if enterprise-owned model keys are used, platform cost is recorded as zero

          Product Screenshots
          SaaS and Independent DeploymentKey

          Same codebase, dual delivery: SaaS out-of-the-box ready, or standalone deployment on enterprise-owned servers

          Same codebase: SaaS out-of-the-box ready; standalone deployment delivers only compiled artifacts, with databases exclusively owned by the client.

          DimensionSaaSIndependent Deployment (Customer On-Premise)
          ServerAIPaddle Cloud ServerYour own Ubuntu server with pm2 + nginx, operating alongside existing systems without Docker.
          DatabaseShared database with multi-tenant row-level security isolationClient-specific Supabase project (single tenant), built from empty database, containing only one tenant and initial administrator
          Scope of featuresAll, including platform operation functionsDuring build, platform operation functions are excluded; no corresponding pages or interfaces exist in output.
          Deliverables—Only compiled artifacts, static resources, runtime dependencies, and environment configurations; zero source code, zero internal documentation, with whitelist access control
          Large Language ModelPlatform key or enterprise-provided key.Through AIPaddle model gateway: client machines hold revocable, individually metered tokens—real keys never stored on client devices; custom models can also be self-configured
          PublishCode merge → full-scale testing → automatic deploymentManual confirmation on deployment console → build, trim, validate → migrate → atomic switch → automatic acceptance, rollback on failure
          Database MigrationOur production line executionExecuted locally on client machine; database credentials never leave the client; each migration runs once, as a standalone transaction.
          BackupPlatform backup.Daily automatic backup, retained for 14 days; additional cloud database platform backup
          SafetyOur operations and maintenanceDaily leakage inspection and delivery audit; License binding to machines with offline grace period (planned); contract enforcement
          How to proceed with each version upgrade (same process every time)
          Only versions that have passed full testing in the test environment are allowed to advance
          ① Retrieve versionOnly retrieve submissions that have passed testing
          ② Check environmentClient machine requires 7 mandatory configurations; missing any prevents release
          ③ CropRemove platform operations functionality
          ④ BuildCompile on our own build machine
          ⑤ Access controlNo source code, no internal labels, no operational routing
          ⑥ TransmissionSync to new directory on customer’s machine first, without switching yet
          ⑦ MigrationClient machine runs locally; fails and stops immediately if execution fails
          ⑧ SwitchAtomic replacement and restart
          ⑨ AcceptanceVersion number, login page, operation routing unreachable; rollback on failure
          Expand: Independent deployment implementation steps
          Independent Deployment Implementation Steps

          APreparation

          Our side only, no changes to client machines
          1. Confirm server: Ubuntu 22.04≥2 vCPUs / 8 GB RAM (recommended: 4 vCPUs); disk storage ≥80G, 443 available, accessible to database, WeCom API, and model gateway
          2. Read-only access to port, memory, installed applications, and security policies; generate "Server Status and Installation Configuration Guide" for client confirmation
          3. Create a dedicated database project for the client, executing full migration from a blank slate, setting up a private storage bucket, and configuring login callback domain
          4. Initialize one tenant and one initial administrator; force password change upon first login
          5. Issue customer-specific tokens via Model Gateway; determine domain name and certificate method

          BEnvironment

          Client written authorization, notify security team
          1. Add swap; install Node LTS, pnpm, pm2, nginx, certbot; run as non-root user
          2. Nginx reverse proxy on port 443, with certificate application and automatic renewal
          3. Write environment configuration: database address and key, session encryption key, credential encryption master key, model gateway token, scheduling key
          4. Configure scheduled job; deploy archive sidecar when session archiving needed, keep private key on client machine

          CFirst release

          Deployment console manual confirmation trigger
          1. Publish the nine-step promotion process as described above
          2. Automatic acceptance: version numbers match, login page functions normally, operation routing is unreachable
          3. Review: product and process normal, customer machine has zero source code, functionality synchronized this time indeed exists in product

          DBusiness Configuration

          Client admin operation, with our team accompanying
          1. Initial administrator logs in and changes password; create department members or enable WeCom contact sync
          2. «WeCom Management» Fill in enterprise info, self-built app, address book Secret, callback URL; add trusted IPs and domains in WeCom backend, configure workspace entry
          3. Create intelligent robot (long connection) and bind object in extensions.
          4. Configure MCP, self-hosted models, knowledge bases, AI Employees, and access permissions as required
          5. Conduct UAT per scenario based on acceptance checklist; retain records

          EOperations and maintenance

          Remote support from our side + client cooperation
          1. Daily leakage inspection and delivery audit
          2. Automatic acceptance fails → revert to previous version; migration fails → do not switch code
          3. Daily database backup, retained for 14 days
          4. Upgrade rhythm: R&D → Test Environment Acceptance → Select Upgrade on Deployment Console → Customer Environment

          Clear boundaries for independent deployment in advance

          • The current database uses a customer-exclusive Supabase cloud project (Singapore region); to keep all data within internal network, migrate to self-hosted PostgreSQL + pgvector, which is a planned item
          • Models default to being accessed through AIPaddle Model Gateway (Qwen); customers may also configure OpenAI-compatible model endpoints within the platform
          • Multi-client parameterized deployment dashboard designed but still under review; License machine binding and offline grace period features under planning
          Safety and Access Permissions

          Positions define what to do; roles define what can be viewed

          Separate positions from roles

          Under construction

          A position defines job content; PMO assigns tasks based on position. A role defines access rights: initiator, owner, member, supervisor. Each project has only one owner; supervisors have read-only access at most.

          Four-tier permission for project files

          Under construction

          Set up access permissions at the first-level directory; for fine-grained control beyond WeCom Drive's capability, enforce via platform

          Department data permissions

          Under construction

          Four data scope levels: All / Department and Subordinates / Department Only / Self; managed uniformly via the access permissions configuration panel

          AI Level-based Autonomy

          Already Live

          Reversible actions are executed directly by AI and can be undone with one click within 12 hours; irreversible actions require approval from the responsible manager.

          Approval Center

          Under construction

          Sharing and authorization of AI capabilities require unified approval—visible who was granted access and who currently holds active permissions

          Evaluation data only visible to authorized personnel

          Under construction

          Member work records and KPIs are visible only to project managers, initiators, and administrators—not shared in group chats; full chat history is not collected.

          Product Screenshots
          PART 3How to implement: 90-day leadership initiativeInitiative · Task Inventory · Alleviate Concerns · Judge Timing · Preserve Core Structure · Roadmap
          First step for top leadership

          Advocate for top leadership to personally use AI to write: this is itself a demonstration.

          • Why now?
          • What should we become?
          • How each individual behaves
          • What to do in the next 90 days
          Core Management Team Example conversation
          On-site Practice · Can AI Do This?

          Break down tasks within a department one by one and ask: Can AI handle this?

          Select who is responsible for each task; the AI's potential contribution percentage is calculated in real time on the right. Use your company's actual tasks in the workshop.

          AI can take on (including human + AI)
          0%
          AI-ledHuman + AIHuman-led
          Alleviate concerns

          The employee's inner thought: 'Is AI here to replace me?'

          All-Staff Group · Presentation and Q&A Example conversation
          1It replaces 'task', not 'person'
          2Everyone can become an Agent Builder for their role
          3Transparent rules with open channels
          Top leadership initiative→Core Layer Consensus→Company-wide Presentation→Open registration
          Judge rhythm

          First assess timing: not every enterprise needs immediate full-scale rewriting.

          Fast Market Changes
          Slow Market Changes
          Business Digitalization Level: Low → High
          “Waiting” also requires preparation

          First cultivate AI-literate personnel and clean up data—only then can the platform fully support operations when mature.

          Preserve Core Structure

          Do not overturn existing systems: retain core structure, accelerate each node with AI

          ERP, IPD, and Management Systems Form the Skeleton; AI Is the Muscles.AIPaddle integrates with systems like SAP via API plugins/MCP without modifying the core architecture.
          Roadmap · Xingyun Zhihang + AIPaddle co-pilot

          90 Days: From a Single Initiative to the First Circle Delivering Results

          Weeks 1–2

          Define direction

          • Top leader uses AI to draft the initiative
          • Establish HARO
          • AIPaddle Activation, WeCom Integration
          Weeks 3–6

          Assemble team

          • Three rounds of registration
          • Select 2–3 pilot communities
          • Circle = Project + WeCom Group
          Weeks 7–10

          Running

          • FDE On-site
          • Launch 3–5 Agents per cycle
          • PMO Automatically Runs Daily and Weekly Reports
          Weeks 11–12

          View results

          • Review based on 5 metrics
          • Agents become organizational assets through accumulation and reuse.
          • Decision to Expand Scope
          Immediate action required

          5 actions that can be done this week, none requiring budget approval.

          0 / 5
          Implementation roadmap and current progress

          First establish daily operational closure before advancing to governance and advanced features.

          Already LiveProduction environment availableUnder constructionTest environment is deployed or under development, awaiting real-device validation.In planningRequirement confirmed, scheduled

          Already Live Available

          • WeCom Foundation: Organization sync, group robot, true @· WeCom push
          • Project Workspace: Automatically create space, internal microdisk management, inbox archiving, pending classification distribution, file summarization
          • PMO Foundation: Group binding, role-based task assignment, modify deadline, bidirectional synchronization of execution plan table, weekly/monthly reports
          • Personal Assistant: Create nine types of objects with one sentence, named invocation, Word reports, project folders, voice drag-and-drop, cross-project Q&A
          • AI Capability Platform: Creation of nine object types, secondary editing, safety checks, release approval and authorization, extension, WeCom management
          • Delivery: SaaS vs customer self-deployed, model gateway, daily backup, and data leakage inspection
          • Short video montage: Media library, media analysis, five confirmation cards, sample clips and daily schedule

          Under construction P1

          • Inspection baseline: Expiry warnings, overdue upgrades, status reminders, performance evaluation logs
          • Control rule execution: Approval matrix extension threshold, each notification includes justification; daily report at 18:00 on workdays
          • Member work records· proactive risk follow-up
          • Team: Job roles separated from positions; AI Employees share same role; project-specific job dictionary
          • Mobile workbench· chat archiving integration, Office collaborative editing
          • Governance: Permission configuration panel, full-filter approval center
          • Usage & Billing: My / Organization usage and monthly bill now live on Pinqi; small text indicating test environment consumption in bottom-right corner is live
          • Copilot Page Help · Ask Questions: Test environment is live, integrated with self-iterating issue log

          In planning P1 Follow-up · P2 · P3

          • P1: First version execution plan & risk log, policies as configuration, project health check, meeting facilitation, issue log, acceptance approval gate, KPI
          • Confirmed: Dynamic task assignment, proactive coordination at critical points
          • P2: Dependencies and Gantt charts, plan vs actual correction, change orders, statistical reports
          • P3: Resource scheduling, revenue and profit margin, contract clause extraction, post-mortem and delivery package
          • Platform: Workflow loops / code / manual input nodes, independent deployment of self-built databases, License binding
          • Enterprise-level WeCom Drive Space: Automatically create public space, outputs organized by object, "Public Files" tab (finalized)
          • SAP Connection Package: Predefined OData list, BTP Destination template, controlled write-back, identity propagation
          Conclusion

          Ensure every super individual’s capabilities remain within and grow inside the organization

          Tools evolve over time, but organizational AI assets, collaboration methods, and governance rules remain stable

          Step 1 · IntegrationEnable SaaS or standalone deployment, synchronize WeCom organizational structure, configure eight zones for WeCom management
          Step 2 · AccumulationDepartments use personal assistants to turn frequent tasks into Agents, workflows, and knowledge bases; after approval, they are reused across the organization
          Step 3 · CollaborationAssign a PMO Agent to one key project, transform policies into rules, and assign AI Employees to roles.
          Step 4 · MeasurementReview usage volume, acceptance pass rate, and KPIs to determine which AI capabilities to institutionalize in the next batch