WeChat / Tencent AI Agent Ecosystem: 2025-2026 Comprehensive

WeChat / Tencent AI Agent Ecosystem: 2025-2026 Comprehensive Research

Compiled: 2026-06-24

Scope: WeChat native AI agents, Tencent Yuanbao, Hunyuan model family, Mini Program agents, WeCom agents, WorkBuddy, and the broader Tencent Agent ecosystem.


Table of Contents

  1. Executive Summary
  2. WeChat Native AI Agent (代号"小微")
  3. Tencent Yuanbao (腾讯元宝) & Yuanqi (腾讯元器)
  4. Tencent Hunyuan Foundation Model & Agent Framework
  5. Tencent Cloud ADP (Agent Development Platform) 3.0
  6. The "Buddy" Family: WorkBuddy & CodeBuddy
  7. ima.copilot — Personal Knowledge Agent
  8. WeCom / Enterprise WeChat Agent "大圆"
  9. WeChat Mini Program AI Agent Integration
  10. OpenClaw / ClawBot — WeChat Bot API
  11. Hy-Memory: Agent Long-Term Memory
  12. Key Architectural Patterns
  13. APIs, SDKs & Integration Points
  14. Timeline & Roadmap

Executive Summary

Tencent has undergone a strategic pivot from "AI restraint" to "AI agent-first" across its entire ecosystem in 2025-2026. The company now fields one of the world's most comprehensive AI agent ecosystems spanning:

  • Consumer AI: WeChat native AI agent, Yuanbao (personal assistant), Yuanqi (agent creation platform for WeChat ecosystem)
  • Workplace AI: WorkBuddy (desktop agent workbench), CodeBuddy (developer coding agent), ima.copilot (knowledge agent)
  • Enterprise AI: WeCom agent "大圆", ADP 3.0 (enterprise agent dev platform), WorkBuddy Enterprise
  • Infrastructure: Hunyuan model family (from 0.5B to trillion-parameter MoE), Hy-Memory (long-term memory plugin), OpenClaw/ClawBot (WeChat Bot API)

In Q3 2025, Tencent President Martin Lau explicitly stated WeChat would "eventually launch an AI agent" — marking the official start of the super-app-to-AI-OS transformation. By June 2026, the WeChat AI agent had entered prototype testing, mini-program AI ecosystem access was announced (June 8), and WeCom's "大圆" agent entered internal beta (June 23).


1. WeChat Native AI Agent

代号/Code Name

Referred to internally as the "highest-priority classified project" (最高优先级核心项目). External reports variously call it "微信AI助手" or "小微" (XiaoWei).

Timeline

MilestoneDate
R&D startedBefore H1 2025 (at least)
Prototype testingEarly 2026
Gray-box (灰度) testingMid-2026 (planned)
Full launch (微信 9.0)Q3 2026 (targeted)
Compliance approvalJune 2026 (initiated)

Core Capability

The WeChat AI agent is designed as a conversational entry point in the chat list that can:

  1. Understand user intent via natural language in the chat interface
  2. Call mini programs automatically — ordering rides (Didi), food delivery (Meituan), booking hotels (Ctrip), buying tickets, etc.
  3. Execute end-to-end workflows — from understanding needs to completing transactions via WeChat Pay
  4. No app-switching required — everything stays inside WeChat

Architecture

  • Entry point: Accessible by swiping right from the WeChat main screen, appearing as a conversation in the chat list
  • Mini Program Bridge: Connects to ~millions of mini programs across the WeChat ecosystem — covering ride-hailing, food delivery, ticketing, shopping, etc.
  • Model: Currently testing multiple models including Zhipu, Alibaba, DeepSeek, and WeChat's in-house small models (NOT exclusively Hunyuan initially). WeChat is also developing its own proprietary AI model (internal codename assigned, expected to land in 2026).
  • Ecosystem Power: Leverages WeChat's four ecosystems:
  • Communication & social (data for understanding user intent)
  • Content (Official Accounts, Channels/视频号)
  • Mini Programs (millions of services)
  • Commerce & payment (WeChat Pay)

Ecosystem Access for Developers

On June 8, 2026, WeChat Open Platform announced that all mini-program developers can opt-in to the WeChat AI ecosystem with two modes:

  • Auto Mode (自动模式): Authorize platform to read mini-program source code; AI auto-analyzes pages and can directly operate them
  • Dev Mode (开发模式): Developers build custom integrations; reviewed and approved before AI can invoke
  • Both modes can be simultaneously enabled

First Batch of Partners

Meituan, JD.com, Didi, Ctrip — confirmed as first-wave ecosystem partners (June 2026).


2. Tencent Yuanbao (腾讯元宝) & Yuanqi (腾讯元器)

Yuanbao (元宝) — Consumer AI Assistant

  • Launched: May 30, 2024
  • Underlying model: Hunyuan (multi-modal: vision, voice, text) + DeepSeek-R1 (added Feb 2025)
  • Key Features:
  • AI Search (integrates WeChat Official Accounts + Channels as information sources)
  • AI Summarization & Writing
  • Document processing (PDF, Word, PPT, Excel, TXT; up to 50 files / 100MB)
  • AI image generation & editing
  • AI video generation (via HunyuanVideo 1.5, Nov 2025)
  • Voice cloning for personalized agents
  • Multi-device: Mobile (iOS/Android), Desktop (Windows/macOS), Web
  • Personal Agent creation — users can create custom agents with personality, voice, and permissions
  • "元宝派" (Yuanbao Circle): One-click access to shared agents like "龙虾" (OpenClaw)
  • "元宝高考通": First industry-specific college admissions counselor agent (with QQ Browser)
  • Integration: Connected with Tencent Docs (Mar 2025), QQ Browser

Yuanqi (元器) — Agent Creation & Distribution Platform

  • Positioning: One-stop agent creation platform for the WeChat ecosystem
  • Low-code/No-code: Create agents in minutes without programming
  • Key Capabilities:
  • Agent types: Public account agents (公众号智能体), customer service agents, knowledge base agents
  • Knowledge base: Import from Official Account articles, documents, custom data
  • Plugin system: Extend agent capabilities via plugins
  • Workflow editor: Visual workflow orchestration
  • Distribution channels: WeChat Official Accounts, Mini Programs, H5, and now Xiaomi App Store (Nov 2025 partnership — first external phone manufacturer)
  • Agent marketplace: Pre-built agents available; community sharing
  • Model support: Hunyuan, DeepSeek, and third-party models
  • Ecosystem position: Positioned for WeChat + WeCom private domain, customer service, and retail scenarios

Yuanbao + Yuanqi Relationship

  • Yuanbao = Consumer-facing AI assistant app (end-user)
  • Yuanqi = Developer/creator platform for building agents that can be published into WeChat ecosystem (developer/business)
  • Agents created on Yuanqi can be embedded in Official Accounts, Mini Programs, and shared via Yuanbao

3. Tencent Hunyuan Foundation Model

Model Family (2025-2026)

ModelParametersKey FeaturesDate
Hunyuan-Large (Open Source)3890B total / 52B activatedMoE architecture, 8×52B experts, 256K context2025
Hunyuan Turbo S-Fast-thinking, general tasksFeb 2025
Hunyuan T1-Deep reasoning, complex tasks2025
Hunyuan 2.0 Think-Enhanced Agent ability, reasoning, 128K input / 64K outputNov 2025
Hunyuan 2.0 Instruct-Instruction following, long-form, code, knowledge accuracyNov 2025
Hunyuan 3.0TBDReduced activation params, improved Agent ability, long memory, multi-turnIn development (Mar 2026 announcement)
Small models (Open Source)0.5B, 1.8B, 4B, 7BConsumer-grade GPU, edge deployment, vertical fine-tuningAug 2025
Hunyuan3D-2.0-3D generation, FlashVDM acceleration frameworkMar 2025
HunyuanVideo 1.58.3BVideo generation, 5-10s HD outputNov 2025
Hunyuan-MT-7B7BTranslation, won 30/31 language pairs at WMT20252025

Agent-Specific Capabilities

  • Hunyuan 2.0: Explicitly enhanced for Agent scenarios — tool calling, multi-step reasoning, complex instruction following
  • Hunyuan 3.0 (upcoming): Focused on long memory, improved Agent ability, multi-turn follow-up
  • API Access: Available via Tencent Cloud API (paid); model names like hunyuan-2.0-thinking-20251109
  • Open Source: Hunyuan-Large, small models (0.5B-7B), Hunyuan3D-2.0, Hunyuan-MT-7B all open-sourced on GitHub/HuggingFace

4. Tencent Cloud ADP 3.0

Agent Development Platform (ADP 3.0)

  • Launched: September 17, 2025 (at Tencent Global Digital Ecosystem Summit)
  • Global availability: Open worldwide
  • Four Major Upgrades:
  1. RAG (Retrieval-Augmented Generation): Enhanced knowledge base management, document conflict comparison, custom chunking, multi-database access, improved table Q&A accuracy. Upgraded from traditional RAG → "Agentic RAG"
  2. Multi-Agent Collaboration: Support for multiple agents working together on complex tasks
  3. WorkFlow Engine: Visual workflow orchestration for complex business processes
  4. Application Building Toolchain: Full lifecycle — dev, debug, evaluate, publish, operate

Architecture Layers

  • Bottom layer: Permission systems, databases, compliance, runtime environments
  • Middle layer: RAG engine, Workflow engine, Agent engine
  • Top layer: Model marketplace (Ti-ONE), Plugin factory, Building templates

Model Support

  • Built-in: Hunyuan, DeepSeek
  • Third-party: Via API (compatible with standard protocols)
  • Model Square (Ti-ONE): One-click authorization to call any desired model

Enterprise Features

  • Dual-layer permission system: Enterprise-level + workspace-level access control, down to knowledge-base entry granularity
  • IP Agent: Combined with digital human capabilities for knowledge assets
  • Vertical Agent templates: Quality inspection, media content processing, marketing growth, intelligent inspection
  • Open source commitment: Tencent YouTu Lab key agent technologies will continue to be open-sourced

AI Dev Kit (April 2025)

Tencent Cloud also launched a lighter "AI Dev Kit":

  • Zero-code mode: Pick model → configure prompt → enable web search → one-click deploy to Mini Program / H5 / Official Account customer service — 5 minutes from console to live agent
  • Code mode: Node.js framework for professional customization
  • MCP hosting: Built-in MCP Server hosting
  • Supports DeepSeek + Hunyuan dual engines

5. The "Buddy" Family

Tencent's "Buddy" family was formally introduced at the June 5, 2026 AI Industry Application Conference, covering the workplace productivity spectrum:

WorkBuddy — Desktop Agent Workbench

  • Positioning: "Full-scenario workplace AI agent desktop workbench" (全场景职场AI智能体桌面工作台)
  • Launched: Internal beta Feb 6, 2026; International edition May 28, 2026; Enterprise edition June 5, 2026
  • Developed by: Tencent Cloud CodeBuddy team
  • Core philosophy: "A digital colleague that can actually DO work" — not just chat, but execute
  • Key Capabilities:
  • Natural language tasking: One sentence to describe a task; WorkBuddy autonomously plans and executes
  • Multi-step complex tasks: Automatically decomposes tasks, plans steps, executes operations
  • Local file operations: Reads authorized folders, batch processes files
  • Multi-modal: Documents, spreadsheets, PPTs, data analysis, image processing, voice-to-text
  • Content creation: Reports, articles, presentations, code generation
  • Multi-Agent parallel processing: Multiple agents working simultaneously on different sub-tasks
  • Built-in MCP Server & Skills ecosystem: Compatible with OpenClaw Skills
  • Multi-model support: DeepSeek, Kimi, GLM, MiniMax, etc. — intelligent model selection per task
  • Communication Channels:
  • WeChat (personal + Official Account customer service)
  • Enterprise WeChat (WeCom)
  • Feishu (Lark)
  • DingTalk
  • QQ
  • Distinguishing Features vs Competitors:
  • "免部署·安装即用" — No deployment needed, install and use immediately
  • OpenClaw Skills compatibility — can reuse existing Skill ecosystem
  • Local execution — data stays on user's machine (enterprise security)
  • Remote control via IM — send tasks from phone's WeChat/WeCom, WorkBuddy executes on PC
  • "龙虾" (OpenClaw) nickname: Sometimes called "腾讯版小龙虾" as the Tencent-edition OpenClaw
  • WorkBuddy Enterprise: Released June 2026, includes Agent Suite for team-level AI deployment — "from super individual to super team"

CodeBuddy — Developer Coding Agent

  • Positioning: AI-powered full-lifecycle programming assistant
  • Three form factors:
  1. IDE Plugin — VS Code, JetBrains, CloudStudio
  2. CodeBuddy IDE — Standalone AI-first IDE (public beta Sept 2025)
  3. CodeBuddy Code CLI — Terminal-based AI coding agent (released Sept 9, 2025)
  • CLI Highlights:
  • npm install: npm install -g @tencent-ai/codebuddy-code
  • Natural language to code: "refactor all components in src/ to React Hooks"
  • Covers full DevOps lifecycle
  • 90%+ code generated by AI in internal usage
  • CodeBuddy 2.0 (2026): Open SDK, plugin marketplace, ACP protocol for standardized agent capabilities, enterprise sandbox deployment, private deployment option
  • Internal impact: 4-engineer team delivered 79 versions in 58 days, 90% code AI-generated; 16%+ internal dev efficiency improvement
  • WeChat Mini Program specialization: Special CodeBuddy variant optimized for WeChat Mini Program development — Craft agent, code completion, unit testing, code review, MCP compatibility
  • Models: Hunyuan + DeepSeek dual-engine; international supports GPT & Gemini

Other Buddy Family Members

  • Ardot: Creative design agent
  • Miora (腾讯妙境): Creative & design tool
  • Shared account system & Credits pool across CodeBuddy + WorkBuddy

6. ima.copilot

Personal Knowledge Agent

  • Launched: October 23, 2024 (Mac), Nov 15, 2024 (Windows), Feb-Mar 2025 (Mobile)
  • Full name: ima.copilot ("I am a copilot")
  • Core positioning: "A knowledge base that thinks" — AI-powered personal knowledge management
  • Underlying models: Hunyuan + DeepSeek-R1 dual engine
  • Key Features:
  • Search-Read-Write unified workflow (搜·读·写一体)
  • Knowledge base creation: Upload docs, web pages, Official Account articles, images → organized into personal knowledge base
  • Shared knowledge bases: Multi-user collaboration; "知识库广场" for public sharing
  • Smart Q&A: Based on personal knowledge base or whole web
  • Document interpretation: Summarize, extract, mind-map generation
  • Smart writing: Papers, reports, creative content with templates
  • Multi-platform: macOS, Windows, Android, iOS, WeChat Mini Program ("ima知识库"小程序) — real-time sync

Knowledge Agent "copilot" (April 29, 2026)

ima evolved from a "tool" to an "agent partner" with the copilot release:

  • Memory system: 4 modules — Copilot settings, User profile, Long-term memory, Experience & tips
  • Context awareness: Detects what user is viewing (webpage, file, knowledge base, note) and provides contextually relevant assistance
  • Skills ecosystem: Built-in official skill packs; extensible via additional Skills
  • Agent creation: Users can create their own specialized agents

7. WeCom Agent "大圆"

Enterprise WeChat AI Agent

  • Code name: "大圆" (Da Yuan)
  • Status: Internal beta as of June 23, 2026
  • Key Capabilities:
  • Context-native: Doesn't require users to repeat context — automatically understands based on existing WeCom data (group chats, documents, meetings, emails, schedules)
  • Scene-aware: Identifies the current interface user is on and provides relevant responses
  • Access method: Swipe left from WeCom to invoke; always available as an AI assistant "living inside WeCom"
  • Scope: Initially for internal enterprise collaboration; future expansion into customer engagement scenarios (leveraging WeCom's native connection to WeChat)
  • WeCom CLI (Open Source, March 30, 2026):
  • WeCom open-sourced its CLI tool, exposing 7 core product capabilities (messages, calendar, documents, etc.) to AI agents
  • CLI is closer to Agent workflow patterns than traditional REST APIs — reduces token consumption, simplifies integration
  • Enables AI assistants to process messages, schedule meetings, organize documents via natural commands
  • Ecosystem: Partners like 微伴助手 (Weiban) provide AI Agent SCRM solutions for WeCom — AI chat agents with compliance review, batch customer analysis, automated customer journey

8. Mini Program AI Agent Integration

Developer Access

On June 8, 2026, WeChat officially opened AI ecosystem access for all mini-program developers:

Two Integration Modes:

  1. Auto Mode (自动模式):
  • Developer authorizes WeChat to read mini-program source code during submission review
  • Platform auto-analyzes pages and makes them operable by WeChat AI
  • Zero additional development required
  1. Dev Mode (开发模式):
  • Developers build custom AI integrations based on mini-program business logic
  • Must pass platform evaluation and review
  • More control, more flexibility
  • Both modes can be simultaneously enabled
  • Opt-in based; respects developer autonomy

Third-Party Integration Pathways

Dify, Coze, and other agent platforms can connect to WeChat Mini Programs via:

  • HTTP API calls from Mini Program (wx.request)
  • Server-side proxy for authentication and data transformation
  • WeChat customer service message API

Enterprise WeChat App AI

The WeChat Open Platform now supports embedding AI agents into enterprise apps, turning Mini Programs from "lightweight tools" into "intelligent service entry points."


9. OpenClaw / ClawBot — WeChat Bot API

Historical Breakthrough

March 2026: WeChat launched the ClawBot plugin — the first-ever official API for personal WeChat bot integration.

Architecture

  • Protocol: iLink (智联)
  • Domain: ilinkai.weixin.qq.com (Tencent official servers)
  • Framework: OpenClaw (open-source AI Gateway framework)
  • Prerequisites: iOS WeChat 8.0.70+ or Android latest (gradually rolling out; HarmonyOS not yet supported)

Integration Flow

  1. WeChat side: Me → Settings → Plugins → ClawBot → Enable
  2. OpenClaw side: Run npx -y @tencent-weixin/openclaw-weixin-cli@latest install
  3. Pairing: Scan QR code with WeChat; authorization confirmed on phone
  4. Result: ClawBot appears as a chat in WeChat; messages sent to it are processed by OpenClaw's configured AI

Significance

  • First legal personal WeChat Bot API in history
  • No third-party protocol simulation (eliminates account banning risk)
  • Enables arbitrary AI model integration into personal WeChat
  • OpenClaw Skills ecosystem is natively reusable
  • Community alternatives also exist: agent-wechat, wechat-agent-channel

OpenClaw as Gateway

OpenClaw serves as the AI Gateway orchestrating:

  • Channel management (WeChat, WeCom, Feishu, DingTalk, QQ, etc.)
  • Model routing (multi-model, intelligent selection)
  • Skills/Plugins system
  • Memory management (with Hy-Memory integration)
  • Task execution on local machine

10. Hy-Memory

Agent Long-Term Memory Plugin

  • Released: May 28, 2026, by Tencent Hunyuan team
  • Designed for: OpenClaw and long-term collaborative agents
  • Tagline: "Agent's second brain" (智能体的第二大脑)

Architecture: Three-Layer Design

#### Layer 1: 6-Level Memory Framework

Memories are decomposed into 6 hierarchical levels (not flat vectors):

  1. Atomic Traces — Raw interaction records
  2. Atomic Facts — Extracted factual knowledge
  3. Identity Profile — User characteristics, preferences
  4. Session Summaries — Compressed conversation summaries
  5. Mental Models — Deep understanding of user's thinking patterns
  6. Forward-looking Intentions — Anticipated future needs

→ Query precision: Agent retrieves only the relevant memory tier, not all conversation history

#### Layer 2: Dual System Design (System 1 / System 2)

  • System 1 ("Day Shift"): Real-time memory writing during user interaction — records traces, extracts facts, updates profiles, compresses summaries. Ensures next message can use new information immediately.
  • System 2 ("Night Shift"): Asynchronous background processing — builds mental models, constructs knowledge networks, predicts intentions. Runs at seconds-to-minutes granularity for deep cognitive consolidation.

#### Layer 3: Evolution Chain (演化链)

  • Problem it solves: Traditional systems either overwrite (keep only latest) or accumulate (keep everything, retrieval chaos)
  • Solution: supersedes pointers chain old and new memories. When any node in the chain is hit during search, the entire evolution history is retrieved.
  • Result: Agent understands how user's decisions and preferences have evolved over time

Performance Metrics

MetricResult
LongMemEval benchmark85.20% (highest among comparable frameworks)
vs Mem0Memory count reduced 70%+, information density +45%
vs GraphitiWrite speed 8× faster
Token consumption35% reduction for ultra-long contexts
Memory update speed20% faster

Deployment Options

  • Lite / Pro / Ultra three configurations
  • One-line command integration
  • No external service deployment required

11. Architectural Patterns

1. Super-App-as-Agent-Platform

WeChat is evolving from a social/communication app into an Agent Operating System:

  • Entry: Chat interface (natural language, right-swipe access)
  • Execution layer: Millions of Mini Programs (services)
  • Transaction layer: WeChat Pay (instant completion)
  • Data layer: Social graph, content consumption, behavioral data → intent understanding

2. Multi-Agent Collaboration (ADP 3.0)

  • Multiple specialized agents collaborate on complex tasks
  • Orchestration via Workflow engine
  • Agentic RAG: agents actively evaluate retrieval quality and re-search if needed

3. Planning-Memory-Tool-Use Trinity

Standard across all Tencent agents:

  • Planning: Task decomposition, chain-of-thought, self-reflection
  • Memory: Short-term (session context) + Long-term (Hy-Memory, vector DB)
  • Tool Use: API calls, file operations, Mini Program invocation, MCP protocol

4. Skills / Plugin Ecosystem

  • OpenClaw Skills format is becoming the lingua franca across Tencent's agent products
  • WorkBuddy compatible with OpenClaw Skills
  • ima supports Skill loading
  • SkillHub for community sharing
  • MCP (Model Context Protocol) supported across the ecosystem

5. Multi-Model Orchestration

Rather than betting on a single model, Tencent's agent infrastructure supports:

  • Model routing: Intelligent selection based on task complexity
  • Cost optimization: Simple tasks → cheap models; complex tasks → powerful models
  • Model marketplace: Ti-ONE for model access and management
  • Supported models: Hunyuan family, DeepSeek, GPT, Claude, Gemini, Kimi, GLM, MiniMax, etc.

6. Communication-Channel-Agnostic Agents

Agents can be accessed via:

  • WeChat personal chat / group chat
  • WeCom (Enterprise WeChat)
  • Feishu / Lark
  • DingTalk
  • QQ
  • Web / Desktop / Mobile apps
  • CLI (CodeBuddy Code, WeCom CLI)

7. Human-in-the-Loop (HITL)

  • Critical decision points (transfers, deletions, data modifications) include human approval steps
  • Built into enterprise-grade deployments (ADP, WorkBuddy Enterprise)
  • Compliance review mechanisms (WeCom SCRM agents)

12. APIs, SDKs & Integration Points

Official APIs

API / SDKDescriptionAccess
WeChat ClawBot (iLink)Personal WeChat Bot APIilinkai.weixin.qq.com; requires WeChat 8.0.70+, OpenClaw instance
Hunyuan APIFoundation model access via Tencent Cloudcloud.tencent.com; paid; model names like hunyuan-2.0-thinking-20251109
ADP 3.0 PlatformFull agent development platformTencent Cloud console; global availability
AI Dev Kit5-min agent creation for Mini Program/H5Tencent Cloud; zero-code mode
WeCom CLIOpen-source CLI for WeCom capabilitiesOpen source on GitHub (March 2026)
Yuanqi PlatformAgent creation & distributionyuanqi.tencent.com/agent-shop
CodeBuddy Code CLIAI coding in terminalnpm install -g @tencent-ai/codebuddy-code
CodeBuddy SDKAgent capability integrationVia ACP protocol; plugin marketplace
Hy-MemoryLong-term memory pluginOne-line install for OpenClaw; Lite/Pro/Ultra tiers

Open Source Repositories

  • Hunyuan-Large, Hunyuan-MT-7B, Hunyuan3D-2.0, small models: GitHub + HuggingFace
  • WeCom CLI: Open source
  • OpenClaw: Open-source AI Gateway framework
  • FlashVDM: 3D generation acceleration framework
  • AngelSlim: Model compression toolkit

MCP (Model Context Protocol) Support

  • ADP 3.0 has built-in MCP Server hosting
  • WorkBuddy supports MCP protocol
  • CodeBuddy compatible with MCP ecosystem
  • AI Dev Kit includes MCP Server hosting

Third-Party Integration

  • Dify → WeChat Mini Program via HTTP API
  • Coze → WeChat via Bot API
  • Agent frameworks (LangGraph, CrewAI, AutoGen) → can target Tencent's agent surfaces

13. Timeline & Roadmap

2024
├── May:     Yuanbao (元宝) App launch
├── Sep:     Yuanqi (元器) "公众号智能体" launch
├── Oct:     ima.copilot Mac launch
└── Nov:     ima.copilot Windows launch

2025
├── Jan:     Ma Huateng announces WeChat/QQ AI agents
├── Feb:     Yuanbao adds DeepSeek-R1; Hunyuan Turbo S
├── Mar:     WeCom CLI open-sourced; Hunyuan 3D-2.0 released
├── Aug:     Hunyuan open-sources 4 small models (0.5B-7B)
├── Sep:     ADP 3.0 global launch; CodeBuddy CLI & IDE public beta
├── Nov:     Tencent Q3 earnings: Martin Lau confirms WeChat AI agent
├── Nov:     HunyuanVideo 1.5 + Yuanbao video generation
└── Dec:     Hy-Memory released

2026
├── Jan:     CodeBuddy Code 2.0; Hunyuan 2.0 Think/Instruct
├── Feb:     WorkBuddy internal beta
├── Mar:     WeChat ClawBot plugin launched; WeChat AI agent project exposed
├── Mar:     Hunyuan 3.0 preview (reduced params, enhanced Agent)
├── Apr:     ima copilot knowledge Agent released
├── May:     WorkBuddy International edition
├── May:     Hy-Memory publicly announced (May 28)
├── Jun:     WorkBuddy Enterprise + Agent Suite (June 5)
├── Jun:     WeChat Mini Program AI ecosystem access open (June 8)
├── Jun:     WeCom "大圆" agent internal beta (June 23)
├── Q3:      WeChat 9.0 with built-in super AI agent (targeted)
└── Late:    WeChat own AI model expected to land

Key Takeaways

  1. WeChat is becoming an AI Agent OS. The combination of chat interface, Mini Program ecosystem, WeChat Pay, and social data creates a uniquely powerful agent platform that no other super-app can match in integration depth.
  1. Tencent's agent strategy is multi-layered. From infrastructure (Hunyuan, ADP, Hy-Memory) to consumer (Yuanbao, WeChat agent) to workplace (Buddy family) to enterprise (WeCom, ADP Enterprise) — Tencent covers the full stack.
  1. OpenClaw/ClawBot is the connectivity linchpin. The ClawBot plugin, based on the iLink protocol, provides the first legal pathway for AI agents to operate personal WeChat accounts, and OpenClaw serves as the orchestration gateway across all communication channels.
  1. The "Buddy" family addresses the full workplace spectrum. WorkBuddy for knowledge workers, CodeBuddy for developers, Ardot/Miora for designers — a complete "digital colleague" suite.
  1. Hy-Memory represents a paradigm shift in agent memory. The 6-layer framework + dual system + evolution chain design moves beyond simple vector storage to true cognitive memory management.
  1. Platform openness is accelerating. From ADP 3.0 global launch to WeCom CLI open-source to Mini Program AI ecosystem access — Tencent is actively courting third-party developers and partners.
  1. 2026 Q3 will be the inflection point. With WeChat 9.0, the built-in super agent, and WeCom's "大圆" all converging, Q3 2026 is when AI agents will become a mass-market reality for 1.4 billion WeChat users.

Research compiled from Tencent official announcements, tech media reports, developer documentation, and community sources. All dates and features are as reported as of June 2026.

← 返回案例列表
分享:
🤖 Try Now →
🤖
🎁