WorkBuddy vs Hermes Agent: office product or agent infrastructure?
I recently looked into Tencent WorkBuddy to understand where it fits among the growing number of AI agents. On the surface, it looks similar to Hermes Agent: both can plan multi-step work, use tools, operate files, remember things, run scheduled jobs and respond through messaging apps.
The overlap is real, but the products are aimed at different jobs.
WorkBuddy is a packaged office product. Hermes is agent infrastructure.
I have not used WorkBuddy hands-on, so this comparison is based on its product pages, documentation, pricing and privacy policy rather than a practical benchmark.
What WorkBuddy is trying to be
WorkBuddy describes itself as a "full-scenario AI workbench" for office professionals. You give it a task, its agents plan and execute the work, and it returns completed artifacts rather than stopping at a chat response.[1][2]
Typical examples include:
researching a topic and producing Markdown, Word and PowerPoint reports;
analyzing spreadsheets and creating charts;
drafting emails or weekly reports;
organizing, renaming and converting local files;
working with GitHub, Jira, Gmail, Google Drive, Notion and Slack.[3]
running recurring briefings or monitoring tasks.[5][11]
This is a sensible product direction. Most office workers do not want an agent framework. They want to ask for a report and receive a usable report.
WorkBuddy packages that experience into a desktop and web application with task views, artifact previews, specialized "Experts", a skill marketplace and visual model configuration. It also supports custom model endpoints and Ollama.[6][14]
Tencent's CodeBuddy is the coding-focused product. WorkBuddy expands the same idea into research, documents, data analysis, automation and general office work. The current subscription advertises shared benefits across both products.[1][7]
Where WorkBuddy looks better
WorkBuddy's strongest advantage is productization.
The path from "research this industry" to a report, spreadsheet and presentation is central to the interface. In Hermes, I can assemble the same workflow from research, document, spreadsheet, presentation and delivery skills. That gives me more control, but it also means more moving parts.
For someone who mainly wants to produce office documents, WorkBuddy will probably feel more immediate. The Expert Center and Skill Marketplace also make its capabilities easier to discover. You do not need to understand how an agent is built before using it.
Its permission model is also reasonably thought through. Default Permissions allows normal work inside a task workspace but asks for confirmation before sensitive file operations, deletion, scripts, commands or network-related actions. Full Access removes those prompts and is intended for trusted, recoverable environments.[13]
That is a better default than pretending an autonomous agent cannot damage anything.
Where Hermes is stronger
Hermes is built for people who want to own and extend the agent itself.
It runs across terminal, TUI, desktop, web and messaging platforms. It supports multiple profiles, plugins, MCP servers, custom tools, hooks, local or remote execution backends, delegated agents and an OpenAI-compatible API.[9][17]
The difference becomes obvious with automation. WorkBuddy supports hourly, daily, weekly and one-time tasks, with results sent to Slack, Telegram or Discord.[11] That covers normal personal productivity routines.
Hermes cron goes further. Jobs can be created, edited, paused, resumed and manually triggered. They can load specific skills, run in isolated sessions, deliver to different destinations, perform configuration checks before spending tokens, or run as plain scripts without involving an LLM.[16]
That matters when automation becomes operational rather than convenient. A daily report is one thing. Maintaining services, monitoring failures, updating repositories and preserving a reliable audit trail is another.
Hermes also has a broader messaging gateway. WorkBuddy uses messaging apps mainly as remote controls for the desktop agent, and the computer must remain on.[4] Hermes is designed to live on an always-running server or homelab and can be reached through Telegram, WhatsApp, LINE, WeChat, email, Signal, Teams and other channels.[9]
The memory systems reveal the same difference
WorkBuddy enables memory by default. It extracts useful context from conversations, regenerates summaries every night and lets the user inspect or edit them. It can even import a profile generated by another AI product.[12]
That is convenient and probably the right design for a consumer application.
Hermes memory is more explicit. Its local memory and user profile are bounded and inspectable, while session history and optional external memory providers handle deeper recall.[18] It requires more discipline, but I can see where information lives and decide how it should be retained.
Again, WorkBuddy packages the behavior. Hermes exposes the machinery.
Privacy is the deciding factor for sensitive work
WorkBuddy supports Ollama and says local models can keep model inference on the machine.[6] That is useful, but it does not make the entire application local.
The WorkBuddy Personal privacy policy says prompts, uploaded files and outputs may be processed by Tencent and by third-party model, skill, MCP and integration providers. Personal information is stored on servers in Singapore, and global support or engineering teams, including teams in mainland China, may access it. Model training is disabled by default, and WorkBuddy says it does not use inputs or outputs for training when the user supplies a model through API-key integration.[8]
For ordinary office work, that may be acceptable. I would not use the personal edition for confidential government or client material without a separate legal and security review.
Hermes is open source and can be self-hosted. Its controls include dangerous-command approvals, file-write restrictions, container isolation, MCP credential filtering and session isolation.[10][15]
Self-hosting does not magically make an agent private. If I configure a cloud model, web search provider or external connector, data still leaves the machine. The advantage is that I choose each boundary. I can also use local models and keep the orchestration, memory and files under my control.
Would I replace Hermes with WorkBuddy?
No.
WorkBuddy currently promotes native macOS and Windows clients plus a web application, but not a native Linux build.[1] My own agent runs on an Arch Linux homelab, stays reachable through Telegram and handles scheduled workflows, code, files and system operations. Hermes fits that environment much better.
I also do not want two stateful agents maintaining separate memories, credentials and automations. They would drift, and every duplicated workflow would create another place to debug when something fails.
I can still see a useful role for WorkBuddy. I would try it for rapid PowerPoint, Word or spreadsheet production, provided the material is not sensitive. I would also study its interface because Hermes can learn from the way WorkBuddy presents experts, skills and artifacts to nontechnical users.
My practical split would be:
Hermes remains the system of record, automation engine and general operator.
WorkBuddy is an optional document-production tool.
Similar capabilities do not make two products interchangeable. WorkBuddy removes complexity for office users. Hermes gives technical users control over that complexity. Which one is better depends on whether you want an AI application or an agent platform.
Sources
[1] https://www.workbuddy.ai
[2] https://www.workbuddy.ai/docs/workbuddy/Overview
[3] https://www.workbuddy.ai/docs/workbuddy/From-Beginner-to-Expert-Guide/Function-Description/Connector
[4] https://www.workbuddy.ai/docs/workbuddy/From-Beginner-to-Expert-Guide/Function-Description/Assistant
[5] https://www.workbuddy.ai/docs/workbuddy/From-Beginner-to-Expert-Guide/Practice-Cases/Daily-Briefing
[6] https://www.workbuddy.ai/docs/workbuddy/From-Beginner-to-Expert-Guide/Function-Description/Model
[7] https://www.workbuddy.ai/pricing
[8] https://www.workbuddy.ai/document/privacy-policy
[9] https://hermes-agent.nousresearch.com/docs/llms.txt
[10] https://github.com/NousResearch/hermes-agent
[11] https://www.workbuddy.ai/docs/workbuddy/From-Beginner-to-Expert-Guide/Function-Description/Automation-Guide
[12] https://www.workbuddy.ai/docs/workbuddy/From-Beginner-to-Expert-Guide/Function-Description/Memory
[13] https://www.workbuddy.ai/docs/workbuddy/From-Beginner-to-Expert-Guide/Function-Description/Permission-Modes
[14] https://www.workbuddy.ai/docs/workbuddy/From-Beginner-to-Expert-Guide/Function-Description/Skills-Market
[15] https://hermes-agent.nousresearch.com/docs/user-guide/security
[16] https://hermes-agent.nousresearch.com/docs/user-guide/features/cron
[17] https://hermes-agent.nousresearch.com/docs/user-guide/features/skills
[18] https://hermes-agent.nousresearch.com/docs/user-guide/features/memory
