Tencent WorkBuddy skips the workspace to edit local files in place

Nami Work provisions a dedicated cloud computer for each user to execute office tasks without supervision 24 hours a day, 7 days a week. According to woshipm, users select distinct hardware configurations to match their operational needs. Physical machines on a desk face the same operational reality. Reporting from ifanr notes that a base Mac mini with an M6 chip and 16GB of unified memory starts at 6,999 yuan, while an M5 Pro configuration with 48GB of memory and 2TB of storage costs 25,749 yuan.

Both environments run directly into desktop filesystem chaos. Isolated sandboxes duplicate working directories across remote volumes, breaking local folder sync and forcing humans to resolve the resulting conflicts manually.

The primary operational bottleneck is no longer token generation speed. Sustainable output depends instead on workspace ergonomics: abandoning isolated virtual machines so background tools can write directly to native storage drives.

Mac mini hardware against 24/7 cloud virtual machines

According to an ifanr reviewer, personal computer multitasking is transforming into multi-agent computing where multiple AI agents run background tasks simultaneously. Rather than supervising every prompt, operators now trigger persistent executions. woshipm reported that Nami Work addresses this shift by assigning a cloud computer to each user, running office tasks 24 hours a day, 7 days a week without supervision across selectable hardware configurations.

Physical hardware imposes distinct limits. Apple's M6 chip, its first 2-nanometer Mac processor, raised the entry-level Mac mini price from 4,499 yuan to 6,999 yuan, an increase of more than 50%, sspai reported.

Persistent agents quickly encounter local capacity ceilings. The base 6,999 yuan Mac mini includes 16GB of unified memory and 256GB of storage. That creates immediate headroom problems. Hosting the 33-billion-parameter MiniMax H3 video generation model requires a 498GB Hugging Face download, ifanr noted. Moving to an M5 Pro chip with 48GB of memory and 2TB of storage costs 25,749 yuan, or 1,073 yuan monthly over 24-month installments. Bandwidth reaches 307GB/s here. Even with that speed exceeding the M6 model's 170GB/s, ComfyUI generation took 18 minutes for a 5-second video and nearly 90 minutes for a 15-second clip.

Enterprise demand for local compute persists despite those bottlenecks. An Apple-commissioned Omdia survey of 1,584 enterprise tech leaders and practitioners found that approximately one-third plan to increase on-device AI workloads within one year. Hotpot restaurant chain Haidilao already deployed local Mac mini clusters across nationwide stores to power surveillance and front-of-house decision systems, sspai reported, backed by an upgraded 2.5Gb Ethernet port. Local silicon guarantees control. Yet continuous background agents quickly strain native disks.

AI Execution Architectures: Dedicated Local Nodes vs. Cloud Virtual PCs vs. In-Place Native Agents

Workspace isolation breaks local storage and sync folders

Workspace isolation creates immediate friction when autonomous tools collide with existing folder hierarchies. Claude Code requires conversations to be bound to a workspace, preventing files located outside that workspace from being opened or previewed without additional authorization. This boundary disrupts standard desktop routines. woshipm's authors argue that WorkBuddy improves upon traditional AI agents by bringing AI assistance directly to the user's native file-opening workflow instead of requiring files to be moved into an isolated agent workspace.

Local execution against native directories bypasses sandbox barriers entirely. By targeting files where they already live, local models eliminate the file duplication that fractures cloud sync drives.

Running these tasks locally delivers the responsiveness required for uninterrupted desktop operations. During benchmark testing on an M5 Pro Mac mini, ifanr reported that a 4-bit quantized Qwen3.8 27B model achieved a generation speed of 41.64 tokens per second with a total response time of 2.9 seconds. Oversized weights choke the hardware. On the same machine, a 70B Llama 3.1 model took 2 minutes and 48 seconds to answer a greeting, averaging 0.05 tokens per second. The ifanr reviewer considers models in the range of 20 billion to 30 billion parameters to be the operational sweet spot for an M5 Pro Mac mini with 48GB of unified memory.

Measurements from sspai reinforce the viability of this hardware bracket. In an LM Studio local AI test running a Gemma 4 31B MLX 4-bit model, the M5 Pro Mac mini recorded a median time-to-first-token latency of 0.157 seconds compared to 0.384 seconds on an M4 Pro MacBook Pro. Subsequent generation speeds reached 13.25 tokens/s and 13.20 tokens/s, respectively.

In-place sidebars versus embedded web office suites

Online productivity environments often attempt to solve document modification by pulling the editing canvas entirely into the browser. As reported by woshipm, 360 integrated an Office document editor into its Nami Work platform so users can adjust generated outputs online. Beyond standard Office formats, this web editor accommodates CSV, Markdown, and TXT files. Its slide editor treats text boxes along with image shapes as independent objects that can be clicked and rearranged directly. According to woshipm, Nami Work provides this editing environment free of charge without advertisements or watermarks, bypassing traditional desktop subscriptions like Microsoft Office or WPS that cost several hundred yuan per year.

That browser enclosure creates an ergonomic gap. Every modified asset remains stranded inside an online silo until an operator downloads it, cluttering the local folder structure with redundant copies.

Tencent WorkBuddy counters this isolation by anchoring its interface directly to native storage. According to woshipm, WorkBuddy provides a standalone file browser triggered through the system's right-click context menu. Rather than creating duplicate files across successive downloads, the tool allows direct editing of local Microsoft Word, Excel, and PowerPoint files, saving modifications back to the original local file. It also supports viewing and editing HTML files within its interface. In addition, WorkBuddy's Mini Program version inspects and edits Markdown and HTML files without deducting user points, billing quotas only when an operator queries the AI chat.

Workspace Ergonomics, Local Hardware, and Cloud Execution

• Editing local desktop documents across standard formats like Microsoft Word, Excel, PowerPoint, Markdown, and HTML without generating duplicate files or fragmenting workspace directories. Use Tencent WorkBuddy's context-menu file browser for 'human-machine co-writing' (人机双写), which edits local files in place and auto-saves directly to the source file, bypassing the workspace isolation barriers imposed by tools like Claude Code.
• Running unattended, continuous document generation and processing without tying up local machine resources. Deploy 360 Nami Work, which assigns a cloud computer to each user running 24 hours a day, 7 days a week without supervision, features an auto-save function every 30 seconds, and handles standard Office formats as well as CSV, Markdown, and TXT files.
• Conducting local model inference for agentic tasks in the 20 billion to 30 billion parameter range. Select an M5 Pro Mac mini configured with 48GB of unified memory and 307GB/s bandwidth, which achieves 41.64 tokens per second on a 4-bit quantized Qwen3.8 27B model and a median time-to-first-token latency of 0.157 seconds on Gemma 4 31B MLX. Avoid deploying 70B Llama 3.1 models on this setup, where generation drops to 0.05 tokens per second.
• Executing local generative video pipelines with massive parameter models like the 498GB MiniMax H3 workflow in ComfyUI. Avoid local-only execution on an M5 Pro Mac mini, where rendering a 5-second video takes 18 minutes and a 15-second video takes nearly 90 minutes; offload massive video generation tasks to specialized remote compute instead.
• Balancing hardware expenditure between standard office computing and intensive multi-agent background workloads. Deploy the entry M6 Mac mini with 16GB of unified memory and 256GB of storage starting at 6,999 yuan for standard desktop tasks, but invest in the M5 Pro Mac mini with 48GB of memory and 2TB of storage at 25,749 yuan (or 1,073 yuan per month across 24-month interest-free installments) if hosting multiple concurrent local background agents.

Dual-write concurrency and the mechanics of safe in-place execution

Tencent WorkBuddy structures simultaneous document access through what it terms "human-machine co-writing" (人机双写). It renders Markdown in real time. The interface supports multi-tab file browsing, automatically switching the AI conversation context whenever an operator moves between open file tabs. Users can append model translations or rewrites directly to selected text passages without duplicating files, and processed documents can then be uploaded directly to the ima knowledge base for future reference.

Timed cadences offer an alternative safeguard against write collisions. Nami Work features an auto-save function that saves modifications to the source document every 30 seconds, pairing this with live input previews when users edit Markdown calculation formulas.

Direct native edits change how operators interact with software. woshipm's author argues that using Office in the AI era should require humans only to set goals, exercise judgment, guide aesthetics, and make decisions, delegating remaining execution to AI agents. The application manages the underlying file structure silently in the background.

This operational shift toward local, unattended execution is reflected in enterprise hardware adoption. An Apple-commissioned Omdia survey of 1,584 enterprise tech leaders and practitioners found that Mac adoption in enterprises developing proprietary AI is roughly 1.8 times higher than in enterprises purchasing commercial AI solutions, according to sspai. Large foundation model labs demonstrate the same reliance on native physical environments. As ifanr reported, OpenAI purchased tens of thousands of Mac mini and Mac Studio units to train AI models, while Anthropic rents Mac servers hosted on Amazon AWS.

Workspace segregation creates immediate operational friction during background runs. Claude Code binds sessions to a dedicated workspace, preventing files located outside that perimeter from being opened or previewed without additional authorization. That boundary breaks native habits. According to woshipm's authors, Tencent WorkBuddy avoids this quarantine by operating directly inside the standard file-opening workflow. The tool modifies Microsoft Word, Excel, and PowerPoint documents in place, saving edits back to original local files without generating duplicate copies.

Practitioners configuring local agent hosts should prioritize interfaces built around this model of human-machine co-writing (人机双写). In-place edits protect directory structures and stop background scripts from choking cloud storage. Keep the files where they live.

For readers outside China

• Availability: Tencent WorkBuddy is distributed as a native desktop utility that integrates into system right-click context menus, as well as a WeChat Mini Program. 360 Nami Work operates as a browser-accessible cloud computer platform. Availability outside mainland China, foreign phone registration options, and English-language interfaces are not disclosed in sources. Apple hardware including the M6 and M5 Pro Mac mini is sold internationally, though pricing discussions and channel subsidies cited in the source material reflect the mainland Chinese retail market.
• Pricing: Software options range from freemium utilities to standard recurring suites. 360 Nami Work offers its integrated document editor free of charge without advertisements or watermarks, though specific pricing tiers for its cloud computer hardware configurations are not disclosed in sources. Tencent WorkBuddy allows viewing and editing of Markdown and HTML files in its Mini Program without deducting points, consuming points only during AI chat interactions (monetary costs per point are not disclosed in sources). By comparison, individual annual subscriptions to Microsoft Office or WPS cost several hundred yuan per year. For local hardware, the base M6 Mac mini with 16GB of unified memory and 256GB of storage starts at 6,999 yuan, an increase of more than 50% from the previous generation's 4,499 yuan entry price. An M6 Mac mini with 24GB of memory and 512GB of storage costs 9,999 yuan, an M5 MacBook Air with 24GB of memory and 512GB of storage costs 11,499 yuan, and a high-spec M5 Pro Mac mini with 48GB of memory and 2TB of storage costs 25,749 yuan (or 1,073 yuan per month on a 24-month interest-free installment plan).
• Closest Western equivalents: Claude Code; Microsoft Copilot; Cursor; Amazon WorkSpaces
• Data residency: Data boundaries differ fundamentally by system topology. Tencent WorkBuddy edits native Microsoft Word, Excel, PowerPoint, Markdown, and HTML files directly on local disks without creating copies, though users can optionally upload processed files to Tencent's cloud-based ima knowledge base for cross-session retrieval. 360 Nami Work processes files on a cloud computer running 24 hours a day, 7 days a week, auto-saving document edits to remote storage every 30 seconds; whether customer data remains strictly within mainland China or complies with international privacy frameworks is not disclosed in sources. For fully local setups, running models on an M5 Pro Mac mini via tools like LM Studio keeps all prompt and document data strictly on local storage, a model adopted by commercial operators like Haidilao in on-premise store clusters.

Sources

• ifanr 新 Mac mini 首发实测：我的第一台「多 Agent」电脑 https://ifanr.com/1681211
• woshipm 腾讯WorkBuddy悄悄上新功能，我再也不用把文件搬来搬去了 https://woshipm.com/ai/6475249.html
• sspai 一台主机，多重角色：新款 Mac mini 首发体验 https://sspai.com/post/114829
• woshipm 纳米Work上线Office编辑器，重新定义AI时代的office https://woshipm.com/ai/6476715.html