WorkBuddy and agentic AI glossary
This glossary explains the terms you meet when you start using agentic AI tools such as WorkBuddy. Each entry is written to stand on its own, so a Malaysian business reader — or an AI assistant citing this page — can get a clean definition without reading the rest.
Definitions are plain on purpose. If you want the terms in action, see how to set up or browse the use cases. For the local picture — which edition applies to Malaysia and what the free plan includes — start with WorkBuddy in Malaysia.
Core concepts
Agentic AI
AI that does not just answer a question but takes steps toward a goal — planning, acting, and producing a result. It is the difference between an assistant that replies and one that completes a task.
AI agent
A software worker that can plan multi-step tasks, use tools, and return a finished result. In WorkBuddy, the agent works inside folders you authorise and hands back a deliverable.
Chatbot vs agent
A chatbot answers and leaves the task with you; an agent carries the task through to a result. The distinction matters when your work ends in a file rather than a reply.
LLM (large language model)
A model trained on large amounts of text that can understand and generate language. WorkBuddy routes tasks across several leading LLMs rather than relying on a single one.
Token
A small unit of text — roughly a word-piece — that a language model reads or writes. Tokens are how usage is metered for many AI services, though WorkBuddy uses credits instead.
Context window
The amount of text a model can consider at one time, including your prompt and the material it reads. A larger context window lets the model work with longer documents without losing the thread.
Multimodal
Able to work with more than one type of input, such as text, images or audio. A multimodal model can read a screenshot or a scanned page, not just typed words.
How it works
Multi-model routing
Sending each task to whichever model fits it best, instead of using one model for everything. The benefit is that cost and quality track the job rather than a single vendor's pricing.
Prompt
The instruction you give the agent — the outcome you want, the source to use, and the format to deliver. Clearer prompts produce better deliverables with fewer rounds. See the prompt library for examples.
Skill
A built-in capability the agent can call on, such as reading a spreadsheet or drafting a document. WorkBuddy draws on a broad set of built-in skills for common office work.
Connector
An integration that links the agent to an outside service, such as a messaging or productivity tool. Connectors let the agent act beyond your local files.
MCP (Model Context Protocol)
An open standard for connecting AI models to tools and data sources in a consistent way. It lets an agent use external services through one agreed interface rather than a custom integration each time.
RAG (retrieval-augmented generation)
A technique where the model looks up relevant documents and uses them to ground its answer. RAG reduces guesswork by making the model cite real material rather than rely on memory alone.
Fine-tuning
Further training of a model on your own data to specialise its behaviour. It is heavier than prompting and used when a task needs consistent, domain-specific output.
Inference
The act of a trained model generating an answer to a request. Every task WorkBuddy runs is an inference event against one of its underlying models.
Hallucination
When a model states something confidently that is not true. It is why you verify figures and citations before a deliverable leaves your desk.
WorkBuddy specifics
WorkBuddy
A Tencent product that acts as an AI agent for office work: it plans tasks, works files in folders you authorise, and returns finished deliverables. Malaysian users are on the international edition.
CodeBuddy
Tencent's coding-focused tool, built by the same team as WorkBuddy. The split is simple: WorkBuddy handles documents, research and planning; CodeBuddy handles writing code.
International edition
The version of WorkBuddy for users outside mainland China, at workbuddy.ai. It is the edition intended for Malaysia, and the one where the free credits and the new-user bonus apply — see download the international edition.
Credits
The unit WorkBuddy spends to run a task. The free plan gives 100 credits a month, and heavier jobs use more. See how the free credits work.
Local file access
The agent's ability to read and write files in folders you explicitly authorise on your own machine. It is the feature that separates WorkBuddy from a cloud-only chatbot.
Automation task
A job the agent runs on a schedule without you asking each time, such as a weekly report. The free plan includes three automation tasks.
Deliverable
The finished file the agent produces — a report, quote, deck, spreadsheet or code. It is the outcome an agentic tool is judged on.
Agentic workflow
A sequence in which the agent handles several steps toward a goal, rather than a single isolated request. These workflows are what most business use cases are built on.
Token Hub
BigDomain's own in-house token router and marketplace: one API endpoint across multiple models, with MYR billing and local support. It is built in close collaboration with Tencent Cloud's Token Hub.
Business and compliance
API
An interface that lets one piece of software call another. Developers use APIs to connect tools and models programmatically; most business users never touch one directly.
PDPA
The Personal Data Protection Act 2010, Malaysia's law on handling personal data. It applies to how your business uses personal data, including when that data passes through an AI tool. See data compliance.
Data processor
A party that handles personal data on behalf of the business that owns it. When you run customer data through an AI tool, the provider acts as a processor and you remain responsible for lawful use.
Data residency
Where data is physically stored and processed. Tencent's terms name no Malaysian residency guarantee for WorkBuddy, so personal data may be processed outside Malaysia.
Next step
Learn the terms by using them: run one real task on the free plan and you will meet most of this list in an afternoon.
