Yangyi (@yangyi) is putting NewMax into the crowded desktop AI agent fight with a local-first pitch: let AI read files, call tools, run workflows and automate browser tasks without making sensitive work live by default inside a cloud app.
Yangyi on X The positioning came in a post on X that framed NewMax against Claude Desktop and OpenAI's Codex, with the sharper distinction being privacy and offline execution rather than model quality. NewMax's own homepage is more careful. It does not name Claude Desktop or Codex as direct competitors. It says NewMax connects to OpenAI, Claude, Antigravity, Grok, Kimi, MiniMax, Qwen, DeepSeek, Zhipu, OpenCode, Ollama and local model services, which makes the product less a model replacement than a local execution layer around multiple models.
That distinction matters because NewMax is not trying to win the benchmark race. The product team is betting that the next useful agent interface is closer to a desktop operating surface: files, folders, terminal commands, local Chrome automation, scheduled tasks, Skills, model routing and permission controls in one place. In other words, NewMax is selling workflow custody. The model can come from OpenAI, Anthropic, a Chinese model provider or Ollama. The work, NewMax says, should stay next to the user's files.
The team is selling workflow, not another chat window
NewMax's about page frames the product as built by AI-first practitioners. That emphasis shows up in the product's design: NewMax is organized less like a developer-only coding agent and more like a general-purpose workbench for people who live inside documents, research, media files, reports and browser-based workflows.
The product story on the same page says NewMax began as a late-night idea whose first version ran within a few hours, but that the durable problem was how to bring AI into daily workflows while respecting each user's data, files and pace. The team says it has watched AI's evolution since late 2022 and concluded that AI's value is not making people manage more tools, but pulling them out of repetitive labor.
That is a warmer pitch than the usual agent launch, but the product still has to survive hard questions. NewMax's site does not disclose a legal entity, headquarters, incorporation jurisdiction, funding, revenue, customer count or valuation. It also does not establish whether the product has undergone an independent security audit. The product's security claims should therefore be read as company claims, not third-party verified facts.
What NewMax says the desktop app can do
The core product is a desktop workbench for macOS, Windows and Linux. The download page lists support for Apple silicon and Intel Macs running macOS 12 Monterey or later, Windows 10 64-bit or later, and 64-bit Linux distributions including Ubuntu 22.04+, Debian 12+, Fedora and Arch. It also lists a minimum of 4GB available memory and about 1000MB of disk space.
Inside the product, NewMax says each workspace binds to a local folder and stores conversations, projects, scheduled tasks and Skill configurations separately. The homepage says NewMax can read, summarize, generate and rewrite PDFs, Office files, Markdown, code, SQLite files, images, audio and video, with preview and diff review for file edits. It also claims media operations such as image cropping, compression, format conversion, voiceover, subtitles, background music, video composition and batch export.
The agentic layer is the bigger swing. NewMax says it can call file tools, run terminal commands, search the web, read webpages and coordinate sub-agents. Permission modes, command allowlists and approval controls are the guardrails NewMax says it uses to keep those actions reviewable. The homepage also says NewMax can operate the user's local Chrome to click, fill forms, search, publish and move across pages, then save common flows as reusable browser workflows.
That makes the real comparison more complicated than "Claude Desktop versus NewMax" or "Codex versus NewMax." OpenAI's Codex app is built around agentic work and skills. Anthropic's Claude desktop documentation describes a desktop interface for running multiple Claude Code sessions with an integrated terminal, file editor, visual diff review and live app preview. NewMax is entering a market where the major model companies are already turning chat into work execution.
NewMax's answer is not to own the model. It is to own the local workspace around the model.
The privacy claim is the product
NewMax's security page says all conversations, files, configuration and model preferences are stored on the user's local device, and that NewMax servers do not store user content. It also notes that when users choose cloud models, messages are transmitted to the selected model provider, while bring-your-own API key mode can send messages directly from the user's device to that provider.
The offline claim is central to the pitch. NewMax supports local model services (including Ollama) and can run in a physically disconnected environment, keeping data on-device from generation through processing and storage.
Free personal use, enterprise boundary control
NewMax says the personal edition is free and includes core functions, user-provided API keys, local model support, local files and workspaces, and Skill automation tasks. The enterprise edition is priced by team size and adds member management, unified model configuration, Skill usage statistics, enterprise security and audit, and private deployment consulting.
The enterprise page shows where NewMax likely expects to make money. It pitches private deployment on a customer's own infrastructure, with Docker or Kubernetes support, data kept inside the enterprise boundary, role permissions, SSO, LDAP, OIDC, model allowlists, usage controls, audit logs, support for compliance requirements such as SOC2, and SLA-backed support. The focus is on privacy- and compliance-sensitive teams that need boundary control and auditability.
That commercial shape is familiar: free desktop adoption first, then paid control planes and private deployment for teams that need security review. The hard part is proving that NewMax can be both useful enough for individuals and governable enough for enterprises. Browser automation, terminal access and file editing are the features that make an agent valuable. They are also the features that make security teams ask for logs, permissions, containment, rollback and audit evidence.
NewMax is early enough that many of the facts investors and buyers would normally look for are still missing from public materials. The team has not disclosed funding. It has not named customers. It has not published independent audit results. It has not said how many people are using the product. But the bet is clear: as model companies push agents deeper into the operating system, a local-first wrapper with cross-model routing may appeal to users who want the power of those models without surrendering the whole workspace to one cloud platform.