01
AI Investment &
Proprietary Models
Proprietary Models
Scaling intelligence with long-term R&D, AI infrastructure, and Tencent HYFinancial commitment · Foundation models · Multimodal innovation
TENCENT CLOUD × TELKOM
Executive Technology Briefing · 2026
Tencent’sInnovation in AI
From frontier foundation models to production-grade intelligence for telecom, media, and the digital economy.
01AI Investment & Tencent HY02TokenHub Model Gateway03Agent-Optimized Products04Agent Suite & ClawPro05Cloud Native Build06Tencent FDE Outlook
Telkom Executive ReceptionConfidential · Discussion Draft
TENCENT CLOUD × TELKOM
Tencent Cloud AI Ecosystem
Six layers — from sustained AI investment and frontier models to enterprise outcomes.
02
Model Gateway
Tencent Cloud TokenHubUnified access · Model choice · Intelligent routing · Cost control
03
Agent-Optimized
Products
Products
Reinventing workflows and user experiences around agentsAI-native productivity · Product co-design · Human-in-the-loop
04
Intelligent Agents &
Enterprise Control
Enterprise Control
WorkBuddy · CodeBuddy · ClawProOffice productivity · AI-assisted coding · Enterprise agent management
05
Cloud-Native
Development
Development
Cloud Native Build (CNB)AI-assisted software delivery · Automation · Secure collaboration
06
Future Outlook
Tencent FDE — from technology to measurable outcomesAI migration · Experience transformation · Effective token harnessing
02 / AGENDA
TENCENT · AI COMMITMENT
Tencent Keeps Increasing
Its Investment in AI
A sustained commitment — more compute, stronger models, and more AI products.
79.2RMB BILLION
2025Full-year CapEx
↗
+82% YoY
84.7RMB BILLION
2026 H1CapEx in just six months
THE SIGNAL IS CLEAR
Six months of 2026
Six months of 2026
already surpassed
full-year 2025.
Tencent is continuously stepping up investment in AI infrastructure and innovation.
+176%Q2 CapEx YoY
Sources: Tencent FY2025 Results and Tencent 2026 Interim Report. 2026 figure is H1 actual, not full-year guidance.
03 / INVESTMENT
TENCENT HY · HY4 PREVIEW
A new flagship generation
BUILT FOR
PRODUCTIVITY.
Scaled across model size, context length, and training data — delivering the largest generation-over-generation capability gain measured by Tencent HY and advancing to the open-source frontier.
HY4Preview · Open Source
Total Parameters770B
Mixture-of-Experts flagship
Activated per Token49B
Focused compute at scale
Context Length1M
Long-horizon understanding
Architecture Depth78
Layers in the backbone
Official source: Tencent HY, “Hy4 preview” model card, 28 Aug 2026.
04 / HY4
TENCENT HY
LLM: Tencent HY, The Foundation Model Journey
Long-term commitment to practical AI and innovation
2020–2022
10B–100B-parameter ad recommendation models launched
2021–2022
Trillion-parameter HY NLP model released
2023.04
HY launched one-stop business fine-tuning services
2023.05
Ads, meetings and Tencent Docs adopted HY
2023.09
HY foundation model officially launched
2024.01
HY upgraded to a high-performance MoE model
2024.05
Industry-first Chinese DIT text-to-image modelOpen source
2024.06
HY multimodal model released
2024.09
HY Turbo released
2024.11
HY-largeOpen source
2024.12
HY VideoOpen source
2025.01
HY 3DOpen source
2025.02
HY T1 deep-reasoning preview launched in Tencent Yuanbao
2025.03
Tencent HY MoE flagship models released
Fast thinking: HY Turbos
Deep reasoning: HY T1
Fast thinking: HY Turbos
Deep reasoning: HY T1
2025.03
HY 3D 2.0Open source
2025.04
HY Vision: next-gen multimodal understanding
2025.06–08
HY A13B, 7B–0.5BOpen source
2025.06–11
HY multimodal generation models
HY Image 2.0/3.0 — Open source
HY Video 1.5 — Open source
HY Vision 1.5 — Video generation
HY Image 2.0/3.0 — Open source
HY Video 1.5 — Open source
HY Vision 1.5 — Video generation
2025.12
Tencent HY 2.0 released
2025.07–12
HY World Model 1.0 — Open source
HY 3D Voyager — Open source
HY 3D World 1.1/1.5 — Open source
HY 3D 3.0 — Open source
HY 3D Voyager — Open source
HY 3D World 1.1/1.5 — Open source
HY 3D 3.0 — Open source
2026.01
HY Motion 1.0
3D character animation frameworkOpen source
3D character animation frameworkOpen source
2026.01–02
Translation model wins WMT2025Open source
2026.04
Tencent HY3 previewOpen source
2026.07
Tencent HY3 releasedOpen source
2026.08
Tencent HY4 previewOpen source
Tencent HY
Milestones and release claims reproduced from the supplied reference.
05 / JOURNEY
LLM: HY4 Preview marks a major leap forward, firmly establishing itself in the top tier of global open-source projects
Benchmark figures reproduced from the supplied reference / official HY4 preview model card. Scores use each benchmark’s original metric; higher is better.
06 / BENCHMARKS
Model Gateway · TokenHub
A unified AI model access platform for enterprises and developers
TokenHub is Tencent Cloud’s “one-stop AI large-model service platform”
It aggregates Tencent’s proprietary HY and Youtu models with mainstream third-party models including DeepSeek, MiniMax, Kimi and GLM — providing unified APIs, billing and operations across general chat, deep reasoning, code generation and more.
Lower Integration Cost
One unified API connects multiple mainstream models — no separate integration for each provider.
Improve Development Efficiency
Token Plan plus integrations with mainstream AI coding tools and Claw tools.
Ensure Service Quality
Stable inference, dedicated endpoints and multi-dimensional monitoring.
Protect Platform Security
Access security → transmission security → storage security.
Target Users
App DevelopersEnterprisesAI ProgrammersContent CreatorsOffice Automation
New User BenefitUp to 1M Tokens
per Model
per Model
Free trial quota for new users · valid for 90 days
07 / TOKENHUB
Model Gateway · TokenHub
Whether desktop agents such as WorkBuddy / OpenClaw / AutoClaw, or coding tools such as CodeBuddy Code / Claude Code / Cline / Cursor / OpenCode, everything connects to the same key and the same endpoint: one BaseURL + one API Key. Only the supported protocol differs by tool.
WWorkBuddyTencent desktop agent · OpenAI
OOpenClawOpen-source agent · OpenAI
AAutoClawDesktop agent · OpenAI
>|CodeBuddyTencent AI coding · OpenAI
CCursorAI editor · Anthropic
CClaude CodeAnthropic CLI · Anthropic
/ClineVS Code extension · Dual
OOpenCodeOpen-source CLI · Dual
>>CodexOpenAI CLI · OpenAI
KiKilo CodeLightweight agent · Dual
QQClawTencent series · OpenAI
One Key · One Endpoint
BASE URL · Endpointtokenhub.tencentmaas.com/v1
API KEY · Keysk-xxxx (your TokenHub Key)
THTokenHubOne-stop large-model access · 20+ models · dual-protocol compatibility
OpenAI protocolAnthropic protocolDual protocolTencent ecosystem
1The only platform-specific differences are the
BaseURL and API Key. Switching or invoking a tool only requires replacing these two values.2Reuse the same Key within one product line — one Key can connect all compatible tools, so departments do not need to maintain multiple Keys in the console.
3The protocol is selected by the tool: Claude Code / Cursor use Anthropic; WorkBuddy / OpenClaw / Codex use OpenAI; Cline / OpenCode / Kilo Code support both. TokenHub is compatible with all of them.
08 / TOKENHUB
TENCENT CLOUD · AGENT INFRASTRUCTURE
From model intelligence to production-grade autonomy
Agent-Optimized Cloud Capabilities
One trusted foundation for memory, execution, protection, and observability.
AGENTEnterprise Intelligence CoreThink · Act · Learn · Govern
Select a capability · details open on this page
Agent Capability · 01
TencentDB Agent Memory
L0 → L3 · MEMORY ORGANIZATION
Four-Layer Memory Architecture
Complete evidence at the base, progressively refined into dense and reusable cognition.
L3
Cognitive MemoryUser cognition, task strategies, and reusable high-level understanding
L2
Semantic MemoryConsolidated facts, knowledge patterns, and stable semantic understanding
L1
Episodic MemoryEvents, entities, time, intent, preferences, actions, and outcomes
L0
Raw DialogueOriginal conversations, messages, tool results, and traceable source records
UPWARD ABSTRACTION · HIGHER COGNITIVE DENSITY
Agent Capability · 02
Agent Runtime
CUSTOMER
GROUPS
AGENT
SCENARIOS
GROUPS
AGENT
SCENARIOS
MiniMax · Kimi · Zhipu · Tencent HY
MiniMax · Kimi · Manus · CodeBuddy
Insurance · ECC · CRM SaaS · Education
Agent Runtime
FULL-CHAIN OBSERVABILITY
Agent EngineBuild & Host
Agent GovernGovern & Operate
Agent EvalsEvaluation
Agent MemoryMemory & Intelligence
Agent RegistryAgent Skills
Agent Sandbox
Agent Wall
Ingress ManagementProtocol conversion · sandbox routing · authentication & authorization
Egress ManagementTransparent proxy · policy control · identity injection
Agent Sandbox
Isolated ExecutionVirtualization · network isolation · storage isolation
Multi-EnvironmentCode · mobile · browser · desktop
Lifecycle & StateTriggers · pause/resume · snapshot · branch cloning
Elasticity & EfficiencyMillisecond startup · massive concurrency · large image fleet
Agent IdentityIdentity & Credentials
Agent GatewayModel & Tool Gateway
Agent Capability · 03
LLM WAF
AI-native security control
Protect every prompt, action, and plug-in
Policy enforcement extends from model input and output to tools, tokens, and the agent skill supply chain.
Tencent Cloud LLM WAF adds agent-specific controls beyond conventional content guardrails, protecting the full interaction and execution path.
Injection Defense
Self-developed detection models identify and block multiple prompt-injection patterns.
Compliance Detection
Safety inspection for harmful, illegal, violent, gambling, drug-related, and other restricted content.
Sensitive Data Protection
Enterprise and personal data controls with 50+ built-in rules plus custom regular expressions.
Tool-Call Protection
Detect malicious or unsafe tool invocation with 90+ built-in protection rules.
Compute Guardrails
Coordinate access control through request frequency and token-consumption thresholds.
Supply-Chain Scanning
Detect command injection, exfiltration, prompt injection, and compromised skill dependencies.
Agent Capability · 04
Tencent Cloud LLM Observability · Powered by TCOP
▣Customer Background
As generative AI adoption accelerates, enterprises need to improve visibility into AI systems, reduce operational risk, and control cost.
△Core Pain Points
- Token cost opacity: usage is difficult to trace, causing budget loss.
- Black-box execution: internal logic is invisible and troubleshooting is difficult.
- Security risk: sensitive data and unsafe responses are hard to identify.
- Low operating efficiency: cross-system analysis is manual and model effectiveness lacks objective evaluation.
▤Urgent Needs
Full-Chain VisibilityReal-Time Intelligent AlertsCompliance Audit
Solution Architecture
Standard EditionOne-Line SDK IntegrationTrace / Metrics SupportOpenTelemetry Standard
Business Application (RAG)
Install Tencent Agent Observability SDK
Automatic instrumentation & telemetry export
Automatic instrumentation & telemetry export
TracesMetrics
TCOP LLM / Agent Observability
Monitoring · Analysis · Storage
Monitoring · Analysis · Storage
Intelligent AlertsVisualizationTrace Analysis
◎Core Advantages
One-Line IntegrationNo code changes required; minute-level access.
Complete Trace CoverageEnd-to-end monitoring with no blind spots.
Fine-Grained TracingToken-level analysis for precision positioning.
Dual ObservabilityMetrics monitoring plus trace-level diagnostics.
↗ Customer Value
¥
Cost ReductionPrompt optimization can reduce cost by up to 80%.
↯
Higher EfficiencyToken consumption and failures are located quickly.
✓
ComplianceMeet enterprise-grade governance and audit requirements.
09 / AGENT CAPABILITIES
TENCENT CLOUD · AI-NATIVE PRODUCTIVITY
WorkBuddy vs CodeBuddy · Official Product Experiences
AI-Native Transformation for Everyone
Two AI-native workspaces designed for different stages of enterprise value creation.
Agentic AI Workspace
WorkBuddy · Business & Office Use Cases
Deep research with cross-source validation and complete reports
Create Word, Excel, PPT, PDF, and multimodal deliverables
Analyze data, generate dashboards, and automate recurring work
Build prototypes and applications from natural-language requirements
Official sources: workbuddy.ai · Tencent Cloud WorkBuddy product documentation
CodeBuddy · Software Delivery Use Cases
Turn product ideas into PRDs, prototypes, and implementation plans
Convert Figma designs into maintainable front-end and back-end code
Complete, explain, refactor, test, debug, and review code
Deliver through IDE, plug-in, and CLI workflows with cloud deployment
Official sources: codebuddy.ai · Tencent Cloud CodeBuddy product documentation
VS
10 / PRODUCT EXPERIENCES
AI-Native Transformation for Everyone
TENCENT CLOUD
WorkBuddy
- Intelligent Desktop AI Secretary
- Full Office Workflow Covered:
- File Reading and Creation
- Multimedia Creation
- Data Processing and Analytics
Who can use it?
Product Managers
Operations
Marketing
Analysts
HR
...
20.97MMonthly VisitsSustained traffic scale
#1PC AI Office ToolResearch Report on the Chinese Office Agent Market 2026
13MDaily Active UsersSustained rapid growth
52Versions since launchAvg. update every 2 days · since March
CodeBuddy
- China’s 1st Fully AI-assisted Coding Partner
- Three Forms to Fit Business Needs:
- IDE Plugin (e.g. VS Code, JetBrains)
- Native IDE
- CLI
Who can use it?
Software Developers
DevOps
ML Engineers
Product Managers
Designers
...
1 License for 2 Products
Content structure recreated from the supplied reference; official WorkBuddy and CodeBuddy product assets used.
11 / AI-NATIVE TRANSFORMATION
TENCENT CLOUD · ENTERPRISE AGENT PLATFORM
OpenClaw Enterprise Edition
ClawPro
Employee Claim Portal · Preview
Tencent Cloud ClawProMy OpenClaw
◉
Your Enterprise OpenClaw
A dedicated AI agent managed by your company, with approved models, channels, knowledge, and token quota.
Activation ProgressReady in minutes
My OpenClawEnterprise Agent · ActiveApproved Model & Channels
✓ Agent ready for work
Enterprise Admin Console · Preview
Model ConfigurationEnterprise Controls
Approved Models
ModelProviderQuotaStatus
HYTencent1M
DeepSeekAPI500K
Enterprise ModelPrivateUnlimited
Token MonitoringLive
80Agents
2.15MInput Tokens
9.7KOutput Tokens
2.16MTotal Tokens
Agent UsageModel CostQuota Control
Employee self-service · Enterprise control plane · Auditable token governance
12 / CLAWPRO
ClawPro
Enterprise Member Management · Give every employee a governed OpenClaw Agent
Enterprise Control Plane
Member Management
Manage enterprise members, access permissions, agent quotas, and token limits.
＋ Add Individually⇧ Bulk Import
⌕ Search by member ID...
MEMBER IDROLESTATUSOPENCLAW LIMITDAILY TOKEN LIMITJOINEDACTIONS
alice@acompany.comAdmin● Active3 / 5100,0002025-01-10Reset Key · Disable · Delete
bob@acompany.comMember● Active1 / 350,0002025-02-15Reset Key · Disable · Delete
carol@acompany.comMember● Active2 / 350,0002025-03-01Reset Key · Disable · Delete
david@acompany.comMember● Disabled0 / 350,0002025-03-20Reset Key · Enable · Delete
eve@acompany.comMember● Active4 / 580,0002025-04-05Reset Key · Disable · Delete
5 enterprise members · Individual and bulk onboarding supported
13 / CLAWPRO
TENCENT CLOUD · CLOUD NATIVE BUILD
CNB NPC Quick Start
From Issue to Pull Request — with an AI Teammate
Mention an NPC in an Issue or PR comment. CNB launches a pipeline, provides repository context, and returns the result in the same conversation.
01
▤
WHAT IS AN ISSUE?
The task and discussion space
An Issue records what needs to be done, why it matters, acceptance criteria, comments, and supporting images.
Requirement contextDiscussion historyImages & evidence
issue.comment@npc
02
✦
WHAT IS AN NPC?
An automated role inside CNB
A virtual AI assistant that can reply, review code, and—in Work Mode—complete development tasks end to end.
System NPCCodeBuddy · ready to useCustom NPCRole + prompt in settings.yml
☑ Work Mode · “Let AI Work”Write code · push branch · create PR · respond to review
TIME-LIMITEDCURRENT REPOMAX: DEVELOPER
03
⑂
WHAT IS A PR?
The reviewable change proposal
A Pull Request packages the code diff, commit history, checks, and reviewer feedback before changes enter the target branch.
Code changes & commitsReview & approvalCI status and feedback
YOU · PR COMMENT
@CodeBuddy Review this PR and fix the findings.pull_request.comment@npc
MENTION→
CREATE / REVIEW→
HOW IT WORKS
1Open an Issue or PRGive the task a durable collaboration context.
→
2Mention @CodeBuddyA new comment triggers the NPC event.
→
3Enable Work ModeGrant temporary, repository-scoped authority.
→
4Review the PRNPC reports progress and iterates on feedback.
Built-in NPC: @CodeBuddy · Custom NPC: @repository-path(role) · Editing an old comment does not retrigger the NPC; add a new mention.
14 / CNB
TENCENT CLOUD · CLOUD NATIVE BUILD
Product Demo
CNB
From Issue intent to PR implementation and CI feedback — an AI-native delivery loop.
VIDEO WALKTHROUGH
▶
CNB Demo Video
The video canvas is ready. Once the video is uploaded, it will be placed here with play/pause, timeline, elapsed time, and volume controls.
00:00 / 00:00
Demo flow: Issue intent → AI implementation → PR review → CI feedback.
15 / CNB
TENCENT CLOUD · FDE OUTLOOK
Field Development Engineering · Migration at Scale
FDE Outlook
Migrating all XLSMART workloads to Tencent Cloud through a phased cloud-to-cloud, on-premises, and big-data transformation program.
XLSMART
60% · Applications — IaaS / PaaS / On-Prem
Touchpoint & WebApps · Websites · Digital channels
Enterprise SupportRPA · HC Apps · BTS Web · SAP
Digital BackendMicroservices · PSB · SDP
IntegrationAPI Gateway · Event Broker · File Transfer
Current estate · AWS · Google Cloud · Huawei Cloud · On-Premises
40% · Data Analytics
Data PlatformData Mart · Data Warehouse · Data Lakehouse
Data ProcessingData Ingestion · ETL Tools
AI FactoryAI / ML · GenAI Studio
Data GovernanceCentralized Data Governance
Huawei · AWS · Snowflake · Azure · Google Cloud
Target New CSPTencent Cloud
C2C Migration
✓ Phase 1 completedCloud-to-cloud workloads · AWS / GCP / Huawei → Tencent Cloud
XX%Cost Cut
4.5 moDuration
7Cutovers
100%Success
MIGRATION SCALE
60+Applications
1,200Microservices
900APIs
15 TBCore Data
Migration
Roadmap
Roadmap
C2CMay 2026 · Completed
On-PremSep 2026 · Completed
Big DataDec 2026 · Planned
16 / FDE Outlook
Field Development Engineering · AI-Powered Migration
From Discovery to OutcomesFDE Outlook
Migration Toolchain — an end-to-end system from infrastructure discovery to verified cloud outcomes.
AWS
SOURCE
SOURCE
01Discovery
⌕What exists?Server · VM · DB
Storage · Network · App
Storage · Network · App
AI CAPABILITYInventory discovery
Agentless scan · CMDB
Agentless scan · CMDB
02Assessment
▤What state?OS · CPU · Memory
Utilization · License · EOL
Utilization · License · EOL
AI CAPABILITYUtilization & sizing
OS / EOL risk
OS / EOL risk
03Dependency
⑂Who depends on whom?App → Web → API
DB → Storage
DB → Storage
AI CAPABILITYFlow-based dependency
App → DB → Storage
App → DB → Storage
047R Strategy
☷Which path?Rehost · Replatform
Refactor · Retain · Retire
Refactor · Retain · Retire
AI CAPABILITY7R recommendation
per-app strategy score
per-app strategy score
05Wave Plan
▥In what order?W1 Low-risk · W2 Std
W3 Core · W4 DB · W5
W3 Core · W4 DB · W5
AI CAPABILITYWave grouping
What · When · Why
What · When · Why
06Migration
♧Move itVM→CVM · DB→DTS
VMware→Cloud
VMware→Cloud
AI CAPABILITYWave-plan handoff
Migration tools · refactoring
Migration tools · refactoring
07Validation
♢Does it work?App · Performance
Business Validation
Business Validation
AI CAPABILITYAI observability
CloudQ bill analysis
CloudQ bill analysis
08Optimize
✣Is it right?Cloud Excellence Index
Six pillars · Multi-cloud
Six pillars · Multi-cloud
AI CAPABILITYCloud architecture
excellence · 6-pillar score
excellence · 6-pillar score
Tencent
Cloud
Cloud
4 TOOL CATEGORIES
A · DiscoveryMigraQ
What exists in the IDC?Lightweight AgentSSH / WMIVMware vCenter · RVToolsServiceNow CMDBZabbix · PrometheusNetwork Flow · Cloud API
B · AssessmentMigraQ
How should it migrate?DiscoveryNormalizationDependencyAssessment7R StrategyCloud Mapping · Wave
C · ExecutionMigraQ + CodeBuddy
How to move it?VMware → CloudPhysical Server → CloudDatabase MigrationStorage MigrationFile MigrationContainer Migration
D · OptimizationCloudQ
Is it right on cloud?SecurityReliabilityPerformanceCostOperational ExcellenceSustainability
WHERE AI FITS
MigraQDiscover → Understand → Assess → Decide → Plan
AI-powered migration advisorMigraQ OrchestrationExecute → Validate
Skill-based workflow orchestrationCloudQChatOps · AIOps · CloudOps · 6-pillar optimization
Cost, performance and architecture excellenceCodeBuddyAnalysis code · AI-assisted code refactoring
Accelerate application modernization17 / FDE Outlook
TENCENT CLOUD × TELKOM
From Intelligence to Impact
Thank You.Let’s Build What’s Next.
Together, we can turn frontier models, enterprise agents, AI-native development, and cloud transformation into measurable outcomes.
Models · HY4
Agents · WorkBuddy
Delivery · CNB + FDE
TENCENT
CLOUD ×
TELKOM
CLOUD ×
TELKOM
AI · AGENTS · CLOUD-NATIVE DELIVERYBUILDING THE INTELLIGENT FUTURE TOGETHER
18 / Closing
@CodeBuddy Help me implement this Issue.